HopeSpeechEDI_CBERT_Multi_Malayalam.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "HopeSpeechEDI_CBERT_Multi_Malayalam.ipynb",
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "elqe9PQrVH6P"
      },
      "source": [
        "### Performing checks for the resources available"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fSELLChxa5su",
        "outputId": "4a1a9528-bb5a-4f25-f9c3-0eab6962d1d5"
      },
      "source": [
        "import torch\r\n",
        "\r\n",
        "# If there's a GPU available...\r\n",
        "if torch.cuda.is_available():    \r\n",
        "\r\n",
        "    # Tell PyTorch to use the GPU.    \r\n",
        "    device = torch.device(\"cuda\")\r\n",
        "\r\n",
        "    print('There are %d GPU(s) available.' % torch.cuda.device_count())\r\n",
        "\r\n",
        "    print('We will use the GPU:', torch.cuda.get_device_name(0))\r\n",
        "\r\n",
        "# If not...\r\n",
        "else:\r\n",
        "    print('No GPU available, using the CPU instead.')\r\n",
        "    device = torch.device(\"cpu\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "There are 1 GPU(s) available.\n",
            "We will use the GPU: Tesla T4\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8SokVowWo3cc",
        "outputId": "74c9c86d-aebe-4bae-9a67-7441d083a9ea"
      },
      "source": [
        "!nvidia-smi"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Sun Feb  7 10:15:54 2021       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 460.39       Driver Version: 418.67       CUDA Version: 10.1     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla T4            Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   58C    P8    11W /  70W |     10MiB / 15079MiB |      0%      Default |\n",
            "|                               |                      |                 ERR! |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K-gbOgflVRxg"
      },
      "source": [
        "### Installing huggingface transformers library"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DbDuxKy0a9Zl",
        "outputId": "7b9e50e2-0d6d-414b-ddb8-aac857bfc75d"
      },
      "source": [
        "!pip install transformers sadice"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: transformers in /usr/local/lib/python3.6/dist-packages (4.2.2)\n",
            "Requirement already satisfied: sadice in /usr/local/lib/python3.6/dist-packages (0.1.3)\n",
            "Requirement already satisfied: importlib-metadata; python_version < \"3.8\" in /usr/local/lib/python3.6/dist-packages (from transformers) (3.4.0)\n",
            "Requirement already satisfied: tokenizers==0.9.4 in /usr/local/lib/python3.6/dist-packages (from transformers) (0.9.4)\n",
            "Requirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.6/dist-packages (from transformers) (4.41.1)\n",
            "Requirement already satisfied: packaging in /usr/local/lib/python3.6/dist-packages (from transformers) (20.9)\n",
            "Requirement already satisfied: filelock in /usr/local/lib/python3.6/dist-packages (from transformers) (3.0.12)\n",
            "Requirement already satisfied: sacremoses in /usr/local/lib/python3.6/dist-packages (from transformers) (0.0.43)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.6/dist-packages (from transformers) (2019.12.20)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from transformers) (2.23.0)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from transformers) (1.19.5)\n",
            "Requirement already satisfied: dataclasses; python_version < \"3.7\" in /usr/local/lib/python3.6/dist-packages (from transformers) (0.8)\n",
            "Requirement already satisfied: torch<2.0.0,>=1.0.0 in /usr/local/lib/python3.6/dist-packages (from sadice) (1.7.0+cu101)\n",
            "Requirement already satisfied: typing-extensions>=3.6.4; python_version < \"3.8\" in /usr/local/lib/python3.6/dist-packages (from importlib-metadata; python_version < \"3.8\"->transformers) (3.7.4.3)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.6/dist-packages (from importlib-metadata; python_version < \"3.8\"->transformers) (3.4.0)\n",
            "Requirement already satisfied: pyparsing>=2.0.2 in /usr/local/lib/python3.6/dist-packages (from packaging->transformers) (2.4.7)\n",
            "Requirement already satisfied: joblib in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (1.0.0)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (1.15.0)\n",
            "Requirement already satisfied: click in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (7.1.2)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (2.10)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (3.0.4)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (2020.12.5)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (1.24.3)\n",
            "Requirement already satisfied: future in /usr/local/lib/python3.6/dist-packages (from torch<2.0.0,>=1.0.0->sadice) (0.16.0)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bPwaYLiRSNPC"
      },
      "source": [
        "## Importing necessary packages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0XdFrhr8SRG0"
      },
      "source": [
        "import pandas as pd\r\n",
        "import numpy as np \r\n",
        "import random\r\n",
        "from sklearn.preprocessing import LabelEncoder\r\n",
        "from sklearn.utils.class_weight import compute_class_weight\r\n",
        "import time\r\n",
        "import datetime\r\n",
        "from sklearn.metrics import classification_report,confusion_matrix\r\n",
        "import random\r\n",
        "import time\r\n",
        "import torch.nn as nn\r\n",
        "from transformers import AutoModel , AutoModelForSequenceClassification, AutoConfig , AutoTokenizer , AdamW ,get_linear_schedule_with_warmup \r\n",
        "import torch\r\n",
        "import matplotlib.pyplot as plt\r\n",
        "import seaborn as sns\r\n",
        "from matplotlib import rc\r\n",
        "from matplotlib.ticker import MaxNLocator\r\n",
        "from sadice import SelfAdjDiceLoss"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dROX3juhVCxC"
      },
      "source": [
        "## Setting some configuration values"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Kzafv4BzVCC9"
      },
      "source": [
        "# Use plot styling from seaborn.\r\n",
        "sns.set(style='darkgrid')\r\n",
        "\r\n",
        "# Increase the plot size and font size.\r\n",
        "sns.set(font_scale=1.5)\r\n",
        "plt.rcParams[\"figure.figsize\"] = (16,12)\r\n",
        "\r\n",
        "# Set the seed value all over the place to make this reproducible.\r\n",
        "seed_val = 42\r\n",
        "\r\n",
        "random.seed(seed_val)\r\n",
        "np.random.seed(seed_val)\r\n",
        "torch.manual_seed(seed_val)\r\n",
        "torch.cuda.manual_seed_all(seed_val)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2v1AIN_OVX4t"
      },
      "source": [
        "## Importing drive into the colaboratory"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Lofimyh2a_tp",
        "outputId": "eec91c43-b626-4ed4-b96a-3d2398e44cca"
      },
      "source": [
        "from google.colab import drive \r\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yOK-M00dVeW2"
      },
      "source": [
        "## Importing dataset directories"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-y-JRPKBbCTv"
      },
      "source": [
        "from os import getcwd , listdir \r\n",
        "from os.path import join\r\n",
        "curr_dir = getcwd()\r\n",
        "drive_dir = join(curr_dir , 'drive','MyDrive')\r\n",
        "dataset_dir = join(drive_dir,'Codalab','HopeSpeech Detection For Equality,DiversityAnd Inclusion','Dataset')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7YU1h8WhmqvK",
        "outputId": "96e32dbc-1309-4d7f-b838-3c57eb7c7348"
      },
      "source": [
        "listdir(dataset_dir)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['english_hope_dev.csv',\n",
              " 'english_hope_train.csv',\n",
              " 'malayalam_hope_dev.csv',\n",
              " 'malayalam_hope_train.csv',\n",
              " 'tamil_hope_first_dev.csv',\n",
              " 'tamil_hope_first_train.csv',\n",
              " 'tamil_hope_test.csv',\n",
              " 'malayalam_hope_test.csv',\n",
              " 'english_hope_test.csv',\n",
              " 'english_hope_test _labels.csv',\n",
              " 'malayalam_hope_test _labels.csv',\n",
              " 'tamil_hope_test _labels.csv',\n",
              " 'malayalam_hope_test_labels.csv']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FUPkZxjLVjLK"
      },
      "source": [
        "## Loading training data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 377
        },
        "id": "joone3H4bINR",
        "outputId": "ec1756c9-93e1-4e0b-9b70-ed85b7ac3495"
      },
      "source": [
        "train_df_path = join(dataset_dir,'malayalam_hope_train.csv')\r\n",
        "# Load the dataset into a pandas dataframe.\r\n",
        "train_df = pd.read_csv(train_df_path)\r\n",
        "\r\n",
        "# Report the number of sentences.\r\n",
        "print('Number of training sentences: {:,}\\n'.format(train_df.shape[0]))\r\n",
        "\r\n",
        "# Display 10 random rows from the data.\r\n",
        "train_df.sample(10)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of training sentences: 8,564\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>7913</th>\n",
              "      <td>Iniyippo ningal pubg kalichalum kuzhappamilla ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6065</th>\n",
              "      <td>ആ ചെങ്ങായിക്കെല്ലും എത്ര നല്ല മൊഞ്ചത്തി പെണ്ണി...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1649</th>\n",
              "      <td>Suryammaa ishanikkaaaa happy wedding anniversa...</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>710</th>\n",
              "      <td>Ente bayi ittekkana shoe muthal jaddy vare ava...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1918</th>\n",
              "      <td>Ayaal entha parayunath ennu ayaalkk tahnne ari...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3617</th>\n",
              "      <td>ചക്ക കുരു പൊട്ടിയോ</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>233</th>\n",
              "      <td>പിന്നെ അമ്മയെയും എന്റെ അന്വേഷണം അറിയിക്കണം</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6927</th>\n",
              "      <td>Helo  iku പകരം എന്തുണ്ട് ?</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2094</th>\n",
              "      <td>Kevin mandane kaanumbhaz santhosh pandittine o...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3114</th>\n",
              "      <td>പൊളിച്ചു bro</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text            label\n",
              "7913  Iniyippo ningal pubg kalichalum kuzhappamilla ...  Non_hope_speech\n",
              "6065  ആ ചെങ്ങായിക്കെല്ലും എത്ര നല്ല മൊഞ്ചത്തി പെണ്ണി...  Non_hope_speech\n",
              "1649  Suryammaa ishanikkaaaa happy wedding anniversa...      Hope_speech\n",
              "710   Ente bayi ittekkana shoe muthal jaddy vare ava...  Non_hope_speech\n",
              "1918  Ayaal entha parayunath ennu ayaalkk tahnne ari...  Non_hope_speech\n",
              "3617                                 ചക്ക കുരു പൊട്ടിയോ  Non_hope_speech\n",
              "233          പിന്നെ അമ്മയെയും എന്റെ അന്വേഷണം അറിയിക്കണം      Hope_speech\n",
              "6927                         Helo  iku പകരം എന്തുണ്ട് ?  Non_hope_speech\n",
              "2094  Kevin mandane kaanumbhaz santhosh pandittine o...  Non_hope_speech\n",
              "3114                                       പൊളിച്ചു bro      Hope_speech"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "w4VW351225YK",
        "outputId": "e437fefb-b57c-43da-9b04-ee7a4f9e1cac"
      },
      "source": [
        "train_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Non_hope_speech    6205\n",
              "Hope_speech        1668\n",
              "not-malayalam       691\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ooOv9QZtnLIF"
      },
      "source": [
        "\r\n",
        "le = LabelEncoder()\r\n",
        "train_df['label'] = le.fit_transform(train_df['label'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "a89Bf4LpnigU",
        "outputId": "7552ab06-152a-4164-c8e2-c4c1884e7c0a"
      },
      "source": [
        "train_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1    6205\n",
              "0    1668\n",
              "2     691\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bvNeR4lTVnK0"
      },
      "source": [
        "### Loading `train_sentences` and `train_labels`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EufQZx2dbKMT"
      },
      "source": [
        "# Get the lists of sentences and their labels.\r\n",
        "train_sentences = train_df['text'].values\r\n",
        "train_labels = train_df['label'].values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wnZSg5f7ZWs_"
      },
      "source": [
        "## Saving Class Names"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "83tTT-wbZa1o",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "724425fd-d546-43da-b77d-388b1d345baf"
      },
      "source": [
        "class_names = le.inverse_transform(range(0,len(train_df['label'].unique())))\r\n",
        "class_names"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['Hope_speech', 'Non_hope_speech', 'not-malayalam'], dtype=object)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ToUbAgZ9ap63"
      },
      "source": [
        "## Loading the validation data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 377
        },
        "id": "Dle8VzoAjOLS",
        "outputId": "c1216e82-f029-45b5-f0b2-a8899a9416f2"
      },
      "source": [
        "val_df_path = join(dataset_dir,'malayalam_hope_dev.csv')\r\n",
        "# Load the dataset into a pandas dataframe.\r\n",
        "val_df = pd.read_csv(val_df_path)\r\n",
        "\r\n",
        "# Report the number of sentences.\r\n",
        "print('Number of validation sentences: {:,}\\n'.format(val_df.shape[0]))\r\n",
        "\r\n",
        "# Display 10 random rows from the data.\r\n",
        "val_df.sample(10)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of validation sentences: 1,070\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>424</th>\n",
              "      <td>ഇന്ത്യൻ മതേതര ജനാധിപത്യ രാജ്യമാണ്</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>722</th>\n",
              "      <td>@Mallu crafts thanik or gay aya kutti ondavath...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>306</th>\n",
              "      <td>Helo. Lesbiann.. chodichal. Vanno</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>116</th>\n",
              "      <td>Nikesh  nte  kada  kettapol  sangadam  തോന്നി...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>551</th>\n",
              "      <td>Ithil parenjekana korey app onnum njan kettit ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>263</th>\n",
              "      <td>Gay enn kandappo nice aayitt kaanan vanirikkka</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>333</th>\n",
              "      <td>Public reactions kalakki but kurach koodi leng...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>575</th>\n",
              "      <td>പോയി ചതൂ ഡേ</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>597</th>\n",
              "      <td>Dr ഈ പ്രതിരോധ ശേഷി ധാരാളം ഉള്ള ആൾക്ക് എത്ര തവണ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>634</th>\n",
              "      <td>Happy mother's day surya amma.. nകരയിപ്പിച്ചു ...</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                  text            label\n",
              "424                  ഇന്ത്യൻ മതേതര ജനാധിപത്യ രാജ്യമാണ്  Non_hope_speech\n",
              "722  @Mallu crafts thanik or gay aya kutti ondavath...  Non_hope_speech\n",
              "306                  Helo. Lesbiann.. chodichal. Vanno  Non_hope_speech\n",
              "116   Nikesh  nte  kada  kettapol  sangadam  തോന്നി...  Non_hope_speech\n",
              "551  Ithil parenjekana korey app onnum njan kettit ...  Non_hope_speech\n",
              "263     Gay enn kandappo nice aayitt kaanan vanirikkka  Non_hope_speech\n",
              "333  Public reactions kalakki but kurach koodi leng...  Non_hope_speech\n",
              "575                                        പോയി ചതൂ ഡേ  Non_hope_speech\n",
              "597  Dr ഈ പ്രതിരോധ ശേഷി ധാരാളം ഉള്ള ആൾക്ക് എത്ര തവണ...  Non_hope_speech\n",
              "634  Happy mother's day surya amma.. nകരയിപ്പിച്ചു ...      Hope_speech"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AypCmtBcDV6v",
        "outputId": "567942cf-0346-4650-e23d-69fc79555c18"
      },
      "source": [
        "val_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Non_hope_speech    784\n",
              "Hope_speech        190\n",
              "not-malayalam       96\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-CgK81lGDPKC"
      },
      "source": [
        "val_df['label'] = le.transform(val_df['label'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Yfgn-E6JDZQZ",
        "outputId": "9dc69d42-f261-4470-f2f8-08a6e9c7c464"
      },
      "source": [
        "val_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1    784\n",
              "0    190\n",
              "2     96\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DMdNNVqrDCuN"
      },
      "source": [
        "### Loading `val_sentences` and `val_labels`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LvGhjUkaDCuO"
      },
      "source": [
        "# Get the lists of sentences and their labels.\r\n",
        "val_sentences = val_df['text'].values\r\n",
        "val_labels = val_df['label'].values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4mgM6b0yL0NM"
      },
      "source": [
        "## Loading the test data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 377
        },
        "id": "d7tVWiIeL0NN",
        "outputId": "7108fd9b-13f1-4455-cfb6-5eafe2ed9d8a"
      },
      "source": [
        "test_df_path = join(dataset_dir,'malayalam_hope_test_labels.csv')\r\n",
        "# Load the dataset into a pandas dataframe.\r\n",
        "test_df = pd.read_csv(test_df_path)\r\n",
        "\r\n",
        "# Report the number of sentences.\r\n",
        "print('Number of test sentences: {:,}\\n'.format(test_df.shape[0]))\r\n",
        "\r\n",
        "# Display 10 random rows from the data.\r\n",
        "test_df.sample(10)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of test sentences: 1,071\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>label</th>\n",
              "      <th>Unnamed: 2</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>673</th>\n",
              "      <td>Sir online വഴി സാധനം വാങ്ങിയാൽ കൊറോണ പകരാൻ ചാൻ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>180</th>\n",
              "      <td>Kundanmar Ennanu Nchangal Vilikkua</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>474</th>\n",
              "      <td>നല്ല മനസ്സുള്ളവർ ഒരുമിച്ചു ജീവിക്കുന്നതല്ലേ മന...</td>\n",
              "      <td>Hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>834</th>\n",
              "      <td>ഇതൊക്കെ രാജ്യം നിരോധിക്കണം</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>668</th>\n",
              "      <td>നായിന്റ മക്കളെ.. പോയി തൂങ്ങി മരിച്ചൂടെ. അവന്റെ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1055</th>\n",
              "      <td>@Vyas Deep brother swarganuragam thettanu yoji...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>I respect you..</td>\n",
              "      <td>not-malayalam</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>448</th>\n",
              "      <td>ഇവർ പുരുഷന്മാർ ആണ്.. nപുരുഷനോട് ആകർഷണം ഉള്ള പു...</td>\n",
              "      <td>Hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>629</th>\n",
              "      <td>@Manikandan Sivan മതത്തിലല്ലടോ nഐൻസ്റ്റീൻ കണ്ട...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>488</th>\n",
              "      <td>@rafid pu അവരുടെ ജീവിതം എങ്ങനെയാടോ തകരുന്നേ വീ...</td>\n",
              "      <td>Hope_speech</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text  ... Unnamed: 2\n",
              "673   Sir online വഴി സാധനം വാങ്ങിയാൽ കൊറോണ പകരാൻ ചാൻ...  ...        NaN\n",
              "180                  Kundanmar Ennanu Nchangal Vilikkua  ...        NaN\n",
              "474   നല്ല മനസ്സുള്ളവർ ഒരുമിച്ചു ജീവിക്കുന്നതല്ലേ മന...  ...        NaN\n",
              "834                          ഇതൊക്കെ രാജ്യം നിരോധിക്കണം  ...        NaN\n",
              "668   നായിന്റ മക്കളെ.. പോയി തൂങ്ങി മരിച്ചൂടെ. അവന്റെ...  ...        NaN\n",
              "1055  @Vyas Deep brother swarganuragam thettanu yoji...  ...        NaN\n",
              "4                                       I respect you..  ...        NaN\n",
              "448   ഇവർ പുരുഷന്മാർ ആണ്.. nപുരുഷനോട് ആകർഷണം ഉള്ള പു...  ...        NaN\n",
              "629   @Manikandan Sivan മതത്തിലല്ലടോ nഐൻസ്റ്റീൻ കണ്ട...  ...        NaN\n",
              "488   @rafid pu അവരുടെ ജീവിതം എങ്ങനെയാടോ തകരുന്നേ വീ...  ...        NaN\n",
              "\n",
              "[10 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wkt73boTL0NO",
        "outputId": "eda182a9-ba57-4c43-c306-736a78c63aeb"
      },
      "source": [
        "test_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Non_hope_speech    776\n",
              "Hope_speech        194\n",
              "not-malayalam      101\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ProewaglL0NP"
      },
      "source": [
        "test_df['label'] = le.transform(test_df['label'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GYR2XokSL0NP",
        "outputId": "645fce7c-39d9-480d-8c57-9f6d0f127d0b"
      },
      "source": [
        "test_df['label'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1    776\n",
              "0    194\n",
              "2    101\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eCCNrKNmL0NQ"
      },
      "source": [
        "### Loading `test_sentences` and `test_labels`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NvBSOGdkL0NR"
      },
      "source": [
        "# Get the lists of sentences and their labels.\r\n",
        "test_sentences = test_df['text'].values\r\n",
        "test_labels = test_df['label'].values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fS2UcOiEVuhm"
      },
      "source": [
        "## Handling class imbalance using `sklearn's compute_class_weight`\r\n",
        "\r\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WL4tP_YkI1mY",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c1f99ee9-96ed-4204-a873-f30c49b6a5b9"
      },
      "source": [
        "\r\n",
        "#compute the class weights\r\n",
        "class_wts = compute_class_weight('balanced', np.unique(train_labels), train_labels)\r\n",
        "\r\n",
        "print(class_wts)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[1.71143086 0.46005909 4.13121081]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "z_WjMTuibsrf"
      },
      "source": [
        "## Helper Functions"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Bef28eHNV-_d"
      },
      "source": [
        "#### 1) Update Interval Function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lBK0ujfEbMir"
      },
      "source": [
        "def good_update_interval(total_iters, num_desired_updates):\r\n",
        "    '''\r\n",
        "    This function will try to pick an intelligent progress update interval \r\n",
        "    based on the magnitude of the total iterations.\r\n",
        "\r\n",
        "    Parameters:\r\n",
        "      `total_iters` - The number of iterations in the for-loop.\r\n",
        "      `num_desired_updates` - How many times we want to see an update over the \r\n",
        "                              course of the for-loop.\r\n",
        "    '''\r\n",
        "    # Divide the total iterations by the desired number of updates. Most likely\r\n",
        "    # this will be some ugly number.\r\n",
        "    exact_interval = total_iters / num_desired_updates\r\n",
        "\r\n",
        "    # The `round` function has the ability to round down a number to, e.g., the\r\n",
        "    # nearest thousandth: round(exact_interval, -3)\r\n",
        "    #\r\n",
        "    # To determine the magnitude to round to, find the magnitude of the total,\r\n",
        "    # and then go one magnitude below that.\r\n",
        "\r\n",
        "    # Get the order of magnitude of the total.\r\n",
        "    order_of_mag = len(str(total_iters)) - 1\r\n",
        "\r\n",
        "    # Our update interval should be rounded to an order of magnitude smaller. \r\n",
        "    round_mag = order_of_mag - 1\r\n",
        "\r\n",
        "    # Round down and cast to an int.\r\n",
        "    update_interval = int(round(exact_interval, -round_mag))\r\n",
        "\r\n",
        "    # Don't allow the interval to be zero!\r\n",
        "    if update_interval == 0:\r\n",
        "        update_interval = 1\r\n",
        "\r\n",
        "    return update_interval"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eMGycwGAWFyO"
      },
      "source": [
        "#### 2) Format time function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c9zVyNQLm-3q"
      },
      "source": [
        "def format_time(elapsed):\r\n",
        "    '''\r\n",
        "    Takes a time in seconds and returns a string hh:mm:ss\r\n",
        "    '''\r\n",
        "    # Round to the nearest second.\r\n",
        "    elapsed_rounded = int(round((elapsed)))\r\n",
        "    \r\n",
        "    # Format as hh:mm:ss\r\n",
        "    return str(datetime.timedelta(seconds=elapsed_rounded))\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EBAo04ENWw3n"
      },
      "source": [
        "#### 3) Make Smart Batches function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nof-AUeoiZc2"
      },
      "source": [
        "def make_smart_batches(text_samples, labels, batch_size,tokenizer,max_input_length):\r\n",
        "    '''\r\n",
        "    This function combines all of the required steps to prepare batches.\r\n",
        "    '''\r\n",
        "\r\n",
        "    print('Creating Smart Batches from {:,} examples with batch size {:,}...\\n'.format(len(text_samples), batch_size))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #   Tokenize & Truncate\r\n",
        "    # =========================\r\n",
        "\r\n",
        "    full_input_ids = []\r\n",
        "\r\n",
        "    # Tokenize all training examples\r\n",
        "    print('Tokenizing {:,} samples...'.format(len(labels)))\r\n",
        "\r\n",
        "    # Choose an interval on which to print progress updates.\r\n",
        "    update_interval = good_update_interval(total_iters=len(labels), num_desired_updates=10)\r\n",
        "\r\n",
        "    # For each training example...\r\n",
        "    for text in text_samples:\r\n",
        "        \r\n",
        "        # Report progress.\r\n",
        "        if ((len(full_input_ids) % update_interval) == 0):\r\n",
        "            print('  Tokenized {:,} samples.'.format(len(full_input_ids)))\r\n",
        "\r\n",
        "        # Tokenize the sample.\r\n",
        "        input_ids = tokenizer.encode(text=text,              # Text to encode.\r\n",
        "                                    add_special_tokens=True, # Do add specials.\r\n",
        "                                    max_length=max_input_length,      \r\n",
        "                                    truncation=True,         # Do Truncate!\r\n",
        "                                    padding=False)           # DO NOT pad.\r\n",
        "                                    \r\n",
        "        # Add the tokenized result to our list.\r\n",
        "        full_input_ids.append(input_ids)\r\n",
        "        \r\n",
        "    print('DONE.')\r\n",
        "    print('{:>10,} samples\\n'.format(len(full_input_ids)))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #      Select Batches\r\n",
        "    # =========================    \r\n",
        "\r\n",
        "    # Sort the two lists together by the length of the input sequence.\r\n",
        "    samples = sorted(zip(full_input_ids, labels), key=lambda x: len(x[0]))\r\n",
        "\r\n",
        "    print('{:>10,} samples after sorting\\n'.format(len(samples)))\r\n",
        "\r\n",
        "    import random\r\n",
        "\r\n",
        "    # List of batches that we'll construct.\r\n",
        "    batch_ordered_sentences = []\r\n",
        "    batch_ordered_labels = []\r\n",
        "\r\n",
        "    print('Creating batches of size {:}...'.format(batch_size))\r\n",
        "\r\n",
        "    # Choose an interval on which to print progress updates.\r\n",
        "    update_interval = good_update_interval(total_iters=len(samples), num_desired_updates=10)\r\n",
        "    \r\n",
        "    # Loop over all of the input samples...    \r\n",
        "    while len(samples) > 0:\r\n",
        "        \r\n",
        "        # Report progress.\r\n",
        "        if ((len(batch_ordered_sentences) % update_interval) == 0 \\\r\n",
        "            and not len(batch_ordered_sentences) == 0):\r\n",
        "            print('  Selected {:,} batches.'.format(len(batch_ordered_sentences)))\r\n",
        "\r\n",
        "        # `to_take` is our actual batch size. It will be `batch_size` until \r\n",
        "        # we get to the last batch, which may be smaller. \r\n",
        "        to_take = min(batch_size, len(samples))\r\n",
        "\r\n",
        "        # Pick a random index in the list of remaining samples to start\r\n",
        "        # our batch at.\r\n",
        "        select = random.randint(0, len(samples) - to_take)\r\n",
        "\r\n",
        "        # Select a contiguous batch of samples starting at `select`.\r\n",
        "        #print(\"Selecting batch from {:} to {:}\".format(select, select+to_take))\r\n",
        "        batch = samples[select:(select + to_take)]\r\n",
        "\r\n",
        "        #print(\"Batch length:\", len(batch))\r\n",
        "\r\n",
        "        # Each sample is a tuple--split them apart to create a separate list of \r\n",
        "        # sequences and a list of labels for this batch.\r\n",
        "        batch_ordered_sentences.append([s[0] for s in batch])\r\n",
        "        batch_ordered_labels.append([s[1] for s in batch])\r\n",
        "\r\n",
        "        # Remove these samples from the list.\r\n",
        "        del samples[select:select + to_take]\r\n",
        "\r\n",
        "    print('\\n  DONE - Selected {:,} batches.\\n'.format(len(batch_ordered_sentences)))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #        Add Padding\r\n",
        "    # =========================    \r\n",
        "\r\n",
        "    print('Padding out sequences within each batch...')\r\n",
        "\r\n",
        "    py_inputs = []\r\n",
        "    py_attn_masks = []\r\n",
        "    py_labels = []\r\n",
        "\r\n",
        "    # For each batch...\r\n",
        "    for (batch_inputs, batch_labels) in zip(batch_ordered_sentences, batch_ordered_labels):\r\n",
        "\r\n",
        "        # New version of the batch, this time with padded sequences and now with\r\n",
        "        # attention masks defined.\r\n",
        "        batch_padded_inputs = []\r\n",
        "        batch_attn_masks = []\r\n",
        "        \r\n",
        "        # First, find the longest sample in the batch. \r\n",
        "        # Note that the sequences do currently include the special tokens!\r\n",
        "        max_size = max([len(sen) for sen in batch_inputs])\r\n",
        "\r\n",
        "        # For each input in this batch...\r\n",
        "        for sen in batch_inputs:\r\n",
        "            \r\n",
        "            # How many pad tokens do we need to add?\r\n",
        "            num_pads = max_size - len(sen)\r\n",
        "\r\n",
        "            # Add `num_pads` padding tokens to the end of the sequence.\r\n",
        "            padded_input = sen + [tokenizer.pad_token_id]*num_pads\r\n",
        "\r\n",
        "            # Define the attention mask--it's just a `1` for every real token\r\n",
        "            # and a `0` for every padding token.\r\n",
        "            attn_mask = [1] * len(sen) + [0] * num_pads\r\n",
        "\r\n",
        "            # Add the padded results to the batch.\r\n",
        "            batch_padded_inputs.append(padded_input)\r\n",
        "            batch_attn_masks.append(attn_mask)\r\n",
        "\r\n",
        "        # Our batch has been padded, so we need to save this updated batch.\r\n",
        "        # We also need the inputs to be PyTorch tensors, so we'll do that here.\r\n",
        "        # Todo - Michael's code specified \"dtype=torch.long\"\r\n",
        "        py_inputs.append(torch.tensor(batch_padded_inputs))\r\n",
        "        py_attn_masks.append(torch.tensor(batch_attn_masks))\r\n",
        "        py_labels.append(torch.tensor(batch_labels))\r\n",
        "    \r\n",
        "    print('  DONE.')\r\n",
        "\r\n",
        "    # Return the smart-batched dataset!\r\n",
        "    return (py_inputs, py_attn_masks, py_labels)\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6EIVfH5-mo_f"
      },
      "source": [
        "### 4) Make Smart Batches ( On Test Set)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ggH8AkAEaV9N"
      },
      "source": [
        "def make_smart_batches_on_test(text_samples, text_ids, batch_size,tokenizer,max_input_length):\r\n",
        "    '''\r\n",
        "    This function combines all of the required steps to prepare batches.\r\n",
        "    '''\r\n",
        "\r\n",
        "    print('Creating Smart Batches from {:,} examples with batch size {:,}...\\n'.format(len(text_samples), batch_size))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #   Tokenize & Truncate\r\n",
        "    # =========================\r\n",
        "\r\n",
        "    full_input_ids = []\r\n",
        "\r\n",
        "    # Tokenize all training examples\r\n",
        "    print('Tokenizing {:,} samples...'.format(len(text_samples)))\r\n",
        "\r\n",
        "    # Choose an interval on which to print progress updates.\r\n",
        "    update_interval = good_update_interval(total_iters=len(text_samples), num_desired_updates=10)\r\n",
        "\r\n",
        "    # For each training example...\r\n",
        "    for text in text_samples:\r\n",
        "        \r\n",
        "        # Report progress.\r\n",
        "        if ((len(full_input_ids) % update_interval) == 0):\r\n",
        "            print('  Tokenized {:,} samples.'.format(len(full_input_ids)))\r\n",
        "\r\n",
        "        # Tokenize the sample.\r\n",
        "        input_ids = tokenizer.encode(text=text,              # Text to encode.\r\n",
        "                                    add_special_tokens=True, # Do add specials.\r\n",
        "                                    max_length=max_input_length,      \r\n",
        "                                    truncation=True,         # Do Truncate!\r\n",
        "                                    padding=False)           # DO NOT pad.\r\n",
        "                                    \r\n",
        "        # Add the tokenized result to our list.\r\n",
        "        full_input_ids.append(input_ids)\r\n",
        "        \r\n",
        "    print('DONE.')\r\n",
        "    print('{:>10,} samples\\n'.format(len(full_input_ids)))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #      Select Batches\r\n",
        "    # =========================    \r\n",
        "\r\n",
        "    # Sort the two lists together by the length of the input sequence.\r\n",
        "    samples = sorted(zip(full_input_ids, text_ids), key=lambda x: len(x[0]))\r\n",
        "\r\n",
        "    print('{:>10,} samples after sorting\\n'.format(len(samples)))\r\n",
        "\r\n",
        "    import random\r\n",
        "\r\n",
        "    # List of batches that we'll construct.\r\n",
        "    batch_ordered_sentences = []\r\n",
        "    batch_ordered_ids = []\r\n",
        "\r\n",
        "    print('Creating batches of size {:}...'.format(batch_size))\r\n",
        "\r\n",
        "    # Choose an interval on which to print progress updates.\r\n",
        "    update_interval = good_update_interval(total_iters=len(samples), num_desired_updates=10)\r\n",
        "    \r\n",
        "    # Loop over all of the input samples...    \r\n",
        "    while len(samples) > 0:\r\n",
        "        \r\n",
        "        # Report progress.\r\n",
        "        if ((len(batch_ordered_sentences) % update_interval) == 0 \\\r\n",
        "            and not len(batch_ordered_sentences) == 0):\r\n",
        "            print('  Selected {:,} batches.'.format(len(batch_ordered_sentences)))\r\n",
        "\r\n",
        "        # `to_take` is our actual batch size. It will be `batch_size` until \r\n",
        "        # we get to the last batch, which may be smaller. \r\n",
        "        to_take = min(batch_size, len(samples))\r\n",
        "\r\n",
        "        # Pick a random index in the list of remaining samples to start\r\n",
        "        # our batch at.\r\n",
        "        select = random.randint(0, len(samples) - to_take)\r\n",
        "\r\n",
        "        # Select a contiguous batch of samples starting at `select`.\r\n",
        "        #print(\"Selecting batch from {:} to {:}\".format(select, select+to_take))\r\n",
        "        batch = samples[select:(select + to_take)]\r\n",
        "\r\n",
        "        #print(\"Batch length:\", len(batch))\r\n",
        "\r\n",
        "        # Each sample is a tuple--split them apart to create a separate list of \r\n",
        "        # sequences and a list of labels for this batch.\r\n",
        "        batch_ordered_sentences.append([s[0] for s in batch])\r\n",
        "        batch_ordered_ids.append([s[1] for s in batch])\r\n",
        "\r\n",
        "        # Remove these samples from the list.\r\n",
        "        del samples[select:select + to_take]\r\n",
        "\r\n",
        "    print('\\n  DONE - Selected {:,} batches.\\n'.format(len(batch_ordered_sentences)))\r\n",
        "\r\n",
        "    # =========================\r\n",
        "    #        Add Padding\r\n",
        "    # =========================    \r\n",
        "\r\n",
        "    print('Padding out sequences within each batch...')\r\n",
        "\r\n",
        "    py_inputs = []\r\n",
        "    py_attn_masks = []\r\n",
        "    py_ids = []\r\n",
        "\r\n",
        "    # For each batch...\r\n",
        "    for (batch_inputs, batch_ids) in zip(batch_ordered_sentences, batch_ordered_ids):\r\n",
        "\r\n",
        "        # New version of the batch, this time with padded sequences and now with\r\n",
        "        # attention masks defined.\r\n",
        "        batch_padded_inputs = []\r\n",
        "        batch_attn_masks = []\r\n",
        "        \r\n",
        "        # First, find the longest sample in the batch. \r\n",
        "        # Note that the sequences do currently include the special tokens!\r\n",
        "        max_size = max([len(sen) for sen in batch_inputs])\r\n",
        "\r\n",
        "        # For each input in this batch...\r\n",
        "        for sen in batch_inputs:\r\n",
        "            \r\n",
        "            # How many pad tokens do we need to add?\r\n",
        "            num_pads = max_size - len(sen)\r\n",
        "\r\n",
        "            # Add `num_pads` padding tokens to the end of the sequence.\r\n",
        "            padded_input = sen + [tokenizer.pad_token_id]*num_pads\r\n",
        "\r\n",
        "            # Define the attention mask--it's just a `1` for every real token\r\n",
        "            # and a `0` for every padding token.\r\n",
        "            attn_mask = [1] * len(sen) + [0] * num_pads\r\n",
        "\r\n",
        "            # Add the padded results to the batch.\r\n",
        "            batch_padded_inputs.append(padded_input)\r\n",
        "            batch_attn_masks.append(attn_mask)\r\n",
        "\r\n",
        "        # Our batch has been padded, so we need to save this updated batch.\r\n",
        "        # We also need the inputs to be PyTorch tensors, so we'll do that here.\r\n",
        "        # Todo - Michael's code specified \"dtype=torch.long\"\r\n",
        "        py_inputs.append(torch.tensor(batch_padded_inputs))\r\n",
        "        py_attn_masks.append(torch.tensor(batch_attn_masks))\r\n",
        "        py_ids.append(torch.tensor(batch_ids))\r\n",
        "    \r\n",
        "    print('  DONE.')\r\n",
        "\r\n",
        "    # Return the smart-batched dataset!\r\n",
        "    return (py_inputs, py_attn_masks, py_ids)\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x6EHYpUbz8fo"
      },
      "source": [
        "### 5) Function for calculating accuracy per class"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Knuw_DOei8RO"
      },
      "source": [
        "def accuracy_per_class(preds, labels , label_dict):\r\n",
        "    \r\n",
        "    preds_flat = np.argmax(preds, axis=1).flatten()\r\n",
        "    labels_flat = labels.flatten()\r\n",
        "\r\n",
        "    for label in np.unique(labels_flat):\r\n",
        "        y_preds = preds_flat[labels_flat==label]\r\n",
        "        y_true = labels_flat[labels_flat==label]\r\n",
        "        print(f'Class: {label_dict[label]}')\r\n",
        "        print(f'Accuracy: {round(100*len(y_preds[y_preds==label])/len(y_true) , 2)}\\n')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sOWk_8lpWO17"
      },
      "source": [
        "### 6) Function for plotting training history"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "R7-FAyfnWM65"
      },
      "source": [
        "def plot_training_history(history):\r\n",
        "  fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(18, 6))\r\n",
        "\r\n",
        "  ax1.plot(history['train_loss'], label='train loss')\r\n",
        "  ax1.plot(history['val_loss'], label='validation loss')\r\n",
        "\r\n",
        "  ax1.xaxis.set_major_locator(MaxNLocator(integer=True))\r\n",
        "#   ax1.set_ylim([-0.05, 1.05])\r\n",
        "  ax1.legend()\r\n",
        "  ax1.set_ylabel('Loss')\r\n",
        "  ax1.set_xlabel('Epoch')\r\n",
        "\r\n",
        "  ax2.plot(history['train_acc'], label='train accuracy')\r\n",
        "  ax2.plot(history['val_acc'], label='validation accuracy')\r\n",
        "\r\n",
        "  ax2.xaxis.set_major_locator(MaxNLocator(integer=True))\r\n",
        "  ax2.set_ylim([-0.05, 1.05])\r\n",
        "  ax2.legend()\r\n",
        "\r\n",
        "  ax2.set_ylabel('Accuracy')\r\n",
        "  ax2.set_xlabel('Epoch')\r\n",
        "\r\n",
        "  fig.suptitle('Training history')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "A7QOTORh_JHa"
      },
      "source": [
        "### 7) Function to check accuracy"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yzRDobQ6MwaR"
      },
      "source": [
        "def check_accuracy(predictions,true_labels):\r\n",
        "  # Combine the results across the batches.\r\n",
        "  predictions = np.concatenate(predictions, axis=0)\r\n",
        "  true_labels = np.concatenate(true_labels, axis=0)\r\n",
        "\r\n",
        "  # Choose the label with the highest score as our prediction.\r\n",
        "  preds = np.argmax(predictions, axis=1).flatten()\r\n",
        "\r\n",
        "  # Calculate simple flat accuracy -- number correct over total number.\r\n",
        "  accuracy = (preds == true_labels).mean()\r\n",
        "\r\n",
        "  return accuracy"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D7TtFkV5_Olz"
      },
      "source": [
        "### 8) Function to evaluate model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TSBNdYHYMA7Q"
      },
      "source": [
        "def eval_model(model,py_inputs, py_attn_masks, py_labels,criterion):\r\n",
        "# Prediction on test set\r\n",
        "  t0 = time.time()\r\n",
        "\r\n",
        "  print('Doing validation on {:,} sentences...'.format(len(py_labels)))\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # Tracking variables \r\n",
        "  predictions , true_labels = [], []\r\n",
        "\r\n",
        "  # Choose an interval on which to print progress updates.\r\n",
        "  update_interval = good_update_interval(total_iters=len(py_inputs), num_desired_updates=10)\r\n",
        "\r\n",
        "  # Measure elapsed time.\r\n",
        "  t0 = time.time()\r\n",
        "  total_val_loss = 0\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # For each batch of training data...\r\n",
        "  for step in range(0, len(py_inputs)):\r\n",
        "\r\n",
        "      # Progress update every 100 batches.\r\n",
        "      if step % update_interval == 0 and not step == 0:\r\n",
        "          # Calculate elapsed time in minutes.\r\n",
        "          elapsed = format_time(time.time() - t0)\r\n",
        "          \r\n",
        "          # Calculate the time remaining based on our progress.\r\n",
        "          steps_per_sec = (time.time() - t0) / step\r\n",
        "          remaining_sec = steps_per_sec * (len(py_inputs) - step)\r\n",
        "          remaining = format_time(remaining_sec)\r\n",
        "\r\n",
        "          # Report progress.\r\n",
        "          print('  Batch {:>7,}  of  {:>7,}.    Elapsed: {:}.  Remaining: {:}'.format(step, len(py_inputs), elapsed, remaining))\r\n",
        "\r\n",
        "      # Copy the batch to the GPU.\r\n",
        "      b_input_ids = py_inputs[step].to(device)\r\n",
        "      b_input_mask = py_attn_masks[step].to(device)\r\n",
        "      b_labels = py_labels[step].to(device)\r\n",
        "    \r\n",
        "      # Telling the model not to compute or store gradients, saving memory and \r\n",
        "      # speeding up prediction\r\n",
        "      with torch.no_grad():\r\n",
        "        logits = model(b_input_ids, b_input_mask)\r\n",
        "\r\n",
        "\r\n",
        "      loss = criterion(logits,b_labels)\r\n",
        "      # Move logits and labels to CPU\r\n",
        "      logits = logits.detach().cpu().numpy()\r\n",
        "      label_ids = b_labels.to('cpu').numpy()\r\n",
        "    \r\n",
        "      # Store predictions and true labels\r\n",
        "      predictions.append(logits)\r\n",
        "      true_labels.append(label_ids)\r\n",
        "\r\n",
        "      total_val_loss += loss.item()\r\n",
        "  \r\n",
        "  val_accuracy = check_accuracy(predictions,true_labels)\r\n",
        "\r\n",
        "  validation_time = format_time(time.time() - t0)\r\n",
        "  \r\n",
        "  # Calculate the average loss over all of the batches.\r\n",
        "  avg_val_loss = total_val_loss / len(py_inputs)     \r\n",
        "    \r\n",
        "\r\n",
        "  # print('Accuracy: {:.3f}'.format(val_accuracy))\r\n",
        "  return (avg_val_loss,val_accuracy,validation_time)\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Lrt_FyJIbC8j"
      },
      "source": [
        "### 9) Function for making predictions on our test dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eylCkH2Zipq5"
      },
      "source": [
        "def get_predictions(py_inputs, py_attn_masks, py_labels):\r\n",
        "\r\n",
        "  print('Predicting labels for {:,} test batches...'.format(len(py_labels)))\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # Tracking variables \r\n",
        "  predictions , true_labels = [], []\r\n",
        "\r\n",
        "  # Choose an interval on which to print progress updates.\r\n",
        "  update_interval = good_update_interval(total_iters=len(py_inputs), num_desired_updates=10)\r\n",
        "\r\n",
        "  # Measure elapsed time.\r\n",
        "  t0 = time.time()\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # For each batch of training data...\r\n",
        "  for step in range(0, len(py_inputs)):\r\n",
        "\r\n",
        "      # Progress update every 100 batches.\r\n",
        "      if step % update_interval == 0 and not step == 0:\r\n",
        "          # Calculate elapsed time in minutes.\r\n",
        "          elapsed = format_time(time.time() - t0)\r\n",
        "          \r\n",
        "          # Calculate the time remaining based on our progress.\r\n",
        "          steps_per_sec = (time.time() - t0) / step\r\n",
        "          remaining_sec = steps_per_sec * (len(py_inputs) - step)\r\n",
        "          remaining = format_time(remaining_sec)\r\n",
        "\r\n",
        "          # Report progress.\r\n",
        "          print('  Batch {:>7,}  of  {:>7,}.    Elapsed: {:}.  Remaining: {:}'.format(step, len(py_inputs), elapsed, remaining))\r\n",
        "\r\n",
        "      # Copy the batch to the GPU.\r\n",
        "      b_input_ids = py_inputs[step].to(device)\r\n",
        "      b_input_mask = py_attn_masks[step].to(device)\r\n",
        "      b_labels = py_labels[step].to(device)\r\n",
        "    \r\n",
        "      # Telling the model not to compute or store gradients, saving memory and \r\n",
        "      # speeding up prediction\r\n",
        "      with torch.no_grad():\r\n",
        "          # Forward pass, calculate logit predictions\r\n",
        "          logits = model(b_input_ids, b_input_mask)\r\n",
        "\r\n",
        "      \r\n",
        "      # Move logits and labels to CPU\r\n",
        "      logits = logits.detach().cpu().numpy()\r\n",
        "      label_ids = b_labels.to('cpu').numpy()\r\n",
        "    \r\n",
        "      # Store predictions and true labels\r\n",
        "      predictions.append(logits)\r\n",
        "      true_labels.append(label_ids)\r\n",
        "      # Combine the results across the batches.\r\n",
        "  predictions = np.concatenate(predictions, axis=0)\r\n",
        "  true_labels = np.concatenate(true_labels, axis=0)\r\n",
        "\r\n",
        "      # Choose the label with the highest score as our prediction.\r\n",
        "  preds = np.argmax(predictions, axis=1).flatten()\r\n",
        "\r\n",
        "  return (preds,true_labels)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eiRNCS-EbZmy"
      },
      "source": [
        "### 10) Function for making predictions on test dataset(without labels)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RIuwhVSmZvkd"
      },
      "source": [
        "def get_predictions_test(py_inputs, py_attn_masks,py_ids):\r\n",
        "\r\n",
        "  print('Predicting labels for {:,} test batches...'.format(len(py_inputs)))\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # Tracking variables \r\n",
        "  predictions , true_labels ,ids = [], [] , []\r\n",
        "\r\n",
        "  # Choose an interval on which to print progress updates.\r\n",
        "  update_interval = good_update_interval(total_iters=len(py_inputs), num_desired_updates=10)\r\n",
        "\r\n",
        "  # Measure elapsed time.\r\n",
        "  t0 = time.time()\r\n",
        "\r\n",
        "  # Put model in evaluation mode\r\n",
        "  model.eval()\r\n",
        "\r\n",
        "  # For each batch of training data...\r\n",
        "  for step in range(0, len(py_inputs)):\r\n",
        "\r\n",
        "      # Progress update every 100 batches.\r\n",
        "      if step % update_interval == 0 and not step == 0:\r\n",
        "          # Calculate elapsed time in minutes.\r\n",
        "          elapsed = format_time(time.time() - t0)\r\n",
        "          \r\n",
        "          # Calculate the time remaining based on our progress.\r\n",
        "          steps_per_sec = (time.time() - t0) / step\r\n",
        "          remaining_sec = steps_per_sec * (len(py_inputs) - step)\r\n",
        "          remaining = format_time(remaining_sec)\r\n",
        "\r\n",
        "          # Report progress.\r\n",
        "          print('  Batch {:>7,}  of  {:>7,}.    Elapsed: {:}.  Remaining: {:}'.format(step, len(py_inputs), elapsed, remaining))\r\n",
        "\r\n",
        "      # Copy the batch to the GPU.\r\n",
        "      b_input_ids = py_inputs[step].to(device)\r\n",
        "      b_input_mask = py_attn_masks[step].to(device)\r\n",
        "      b_ids = py_ids[step].to(device)\r\n",
        "    \r\n",
        "      # Telling the model not to compute or store gradients, saving memory and \r\n",
        "      # speeding up prediction\r\n",
        "      with torch.no_grad():\r\n",
        "          # Forward pass, calculate logit predictions\r\n",
        "          outputs = model(b_input_ids, b_input_mask)\r\n",
        "\r\n",
        "      logits = outputs.logits \r\n",
        "\r\n",
        "      # Move logits and labels to CPU\r\n",
        "      logits = logits.detach().cpu().numpy()\r\n",
        "      b_ids = b_ids.detach().cpu().numpy()\r\n",
        "      # Store predictions\r\n",
        "      predictions.append(logits)\r\n",
        "      ids.append(b_ids)\r\n",
        "      # Combine the results across the batches.\r\n",
        "  predictions = np.concatenate(predictions, axis=0)\r\n",
        "  ids = np.concatenate(ids,axis=0)\r\n",
        "  # Choose the label with the highest score as our prediction.\r\n",
        "  preds = np.argmax(predictions, axis=1).flatten()\r\n",
        "\r\n",
        "  return ids,preds"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4IM8r_ddYZZS"
      },
      "source": [
        "### 11) Confusion Matrix Helper Function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DqNaS8SrBox5"
      },
      "source": [
        "def show_confusion_matrix(confusion_matrix, class_names):\r\n",
        "\r\n",
        "  cm = confusion_matrix.copy()\r\n",
        "\r\n",
        "  cell_counts = cm.flatten()\r\n",
        "\r\n",
        "  cm_row_norm = cm / cm.sum(axis=1)[:, np.newaxis]\r\n",
        "\r\n",
        "  row_percentages = [\"{0:.2f}\".format(value) for value in cm_row_norm.flatten()]\r\n",
        "\r\n",
        "  cell_labels = [f\"{cnt}\\n{per}\" for cnt, per in zip(cell_counts, row_percentages)]\r\n",
        "  cell_labels = np.asarray(cell_labels).reshape(cm.shape[0], cm.shape[1])\r\n",
        "\r\n",
        "  df_cm = pd.DataFrame(cm_row_norm, index=class_names, columns=class_names)\r\n",
        "\r\n",
        "  hmap = sns.heatmap(df_cm, annot=cell_labels, fmt=\"\", cmap=\"Blues\")\r\n",
        "  hmap.yaxis.set_ticklabels(hmap.yaxis.get_ticklabels(), rotation=0, ha='right')\r\n",
        "  hmap.xaxis.set_ticklabels(hmap.xaxis.get_ticklabels(), rotation=30, ha='right')\r\n",
        "  plt.ylabel('True Sign')\r\n",
        "  plt.xlabel('Predicted Sign');"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MBxHY4rMojd-"
      },
      "source": [
        "## Loading the tokenizer using AutoTokenizer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fIlwEvcMb4pb",
        "outputId": "610ebac8-d6d1-4dc6-a17e-0967c77fc7cb"
      },
      "source": [
        "model_name = 'bert-base-multilingual-cased'\r\n",
        "# Load the BERT tokenizer.\r\n",
        "print(f'Loading {model_name} tokenizer...')\r\n",
        "tokenizer = AutoTokenizer.from_pretrained(model_name, do_lower_case=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading bert-base-multilingual-cased tokenizer...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OVs3Jb3Vb89i"
      },
      "source": [
        "## Smart Batching of the training data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cWNt4eOLofmF"
      },
      "source": [
        "lengths = []\r\n",
        "for text in train_sentences:\r\n",
        "  lengths.append(len(text))       "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 427
        },
        "id": "hbdGRA3Bo02O",
        "outputId": "b7b56d6a-18c9-468b-e409-a1e9bf0e490c"
      },
      "source": [
        "\r\n",
        "# Use plot styling from seaborn.\r\n",
        "sns.set(style='darkgrid')\r\n",
        "\r\n",
        "# Increase the plot size and font size.\r\n",
        "sns.set(font_scale=1.5)\r\n",
        "plt.rcParams[\"figure.figsize\"] = (12,6)\r\n",
        "\r\n",
        "plt.scatter(range(0, len(lengths)), lengths, marker=\"|\")\r\n",
        "\r\n",
        "plt.xlabel('Sample Number')\r\n",
        "plt.ylabel('Sequence Length')\r\n",
        "plt.title('Samples BEFORE Tokenizing')\r\n",
        "\r\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9rMhLNhPcZjU"
      },
      "source": [
        "### Tokenizing the sequences"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t7JsLi3EdEje"
      },
      "source": [
        "max_input_length = 400"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "J968o8RJcUEd",
        "outputId": "d6ef1207-3783-4bc1-e838-b46bdeb09729"
      },
      "source": [
        "full_input_ids = []\r\n",
        "labels = []\r\n",
        "\r\n",
        "# Tokenize all training examples\r\n",
        "print('Tokenizing {:,} training samples...'.format(len(train_sentences)))\r\n",
        "\r\n",
        "# Choose an interval on which to print progress updates.\r\n",
        "update_interval = good_update_interval(total_iters=len(train_sentences), num_desired_updates=10)\r\n",
        "\r\n",
        "# For each training example...\r\n",
        "for text in train_sentences:\r\n",
        "    \r\n",
        "    # Report progress.\r\n",
        "    if ((len(full_input_ids) % update_interval) == 0):\r\n",
        "        print('  Tokenized {:,} samples.'.format(len(full_input_ids)))\r\n",
        "\r\n",
        "    # Tokenize the sentence.\r\n",
        "    input_ids = tokenizer.encode(text=text,           \r\n",
        "                                 add_special_tokens=True, \r\n",
        "                                 max_length=max_input_length,  \r\n",
        "                                 truncation=True,     \r\n",
        "                                 padding=False)       \r\n",
        "                                 \r\n",
        "    # Add the tokenized result to our list.\r\n",
        "    full_input_ids.append(input_ids)\r\n",
        "    \r\n",
        "print('DONE.')\r\n",
        "print('{:>10,} samples'.format(len(full_input_ids)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tokenizing 8,564 training samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,800 samples.\n",
            "  Tokenized 2,700 samples.\n",
            "  Tokenized 3,600 samples.\n",
            "  Tokenized 4,500 samples.\n",
            "  Tokenized 5,400 samples.\n",
            "  Tokenized 6,300 samples.\n",
            "  Tokenized 7,200 samples.\n",
            "  Tokenized 8,100 samples.\n",
            "DONE.\n",
            "     8,564 samples\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rZzhO2tZeXur"
      },
      "source": [
        "# Get all of the lengths.\r\n",
        "unsorted_lengths = [len(x) for x in full_input_ids]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 427
        },
        "id": "x7c_mYuVed_H",
        "outputId": "94fa4743-55c9-489f-fd1c-3b1689b0249a"
      },
      "source": [
        "\r\n",
        "# Use plot styling from seaborn.\r\n",
        "sns.set(style='darkgrid')\r\n",
        "\r\n",
        "# Increase the plot size and font size.\r\n",
        "sns.set(font_scale=1.5)\r\n",
        "plt.rcParams[\"figure.figsize\"] = (12,6)\r\n",
        "\r\n",
        "plt.scatter(range(0, len(unsorted_lengths)), unsorted_lengths, marker=\"|\")\r\n",
        "\r\n",
        "plt.xlabel('Sample Number')\r\n",
        "plt.ylabel('Sequence Length')\r\n",
        "plt.title('Samples BEFORE Sorting')\r\n",
        "\r\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAusAAAGaCAYAAAC2bw3EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzde0AU5f4/8PeCgIooXtYbCmgKmAiIlWlleUmJUlBB8UKhRnrU8w3UUjp2TsfyGt7KyrxhEmZpIF6Oty6/OmnHLqikoh5vCZKyotyFRdzfH5zdXHaB2WV2d2Z5v/5RZp6d+czMM7OfefaZZxQajUYDIiIiIiKSHAdbB0BERERERMYxWSciIiIikigm60REREREEsVknYiIiIhIopisExERERFJFJN1IiIiIiKJYrJORGRh0dHRGDJkiK3DIBnLycmBr68v3n//fVuHQkRW1sTWARARGZOdnY0NGzbg559/xh9//AFnZ2e0a9cOAQEBGD16NB5//HFbhygLqampSEhI0JvWrFkzdOjQAU899RSmTZuGTp066ebl5ORg6NChdS7zu+++Q8eOHQEAQ4YMwfXr142WGz9+PBYtWqQ37dy5c0hKSsJPP/0ElUqF5s2bw9fXF+Hh4QgPD4ejo6Ne+ffffx/r1q3T/a1QKNCyZUv06tULL774otFYfX1964w/JSUFjzzySJ1liouLkZycjEOHDuH69euoqqpC69at4efnh8GDByMyMrLOz5sjJycHaWlpGDZsGHr16iX68olInpisE5Hk/Pbbb4iOjkaTJk0QHh6OHj16oLy8HL///juOHj0KV1dXJusmio6ORp8+fQAApaWl+O2335CSkoIjR45g//79aNGihV75J554AmFhYUaX1apVK72/O3bsiDlz5hiU69atm97f27dvxzvvvIOWLVti9OjR6N69O4qKinD48GG88cYb2L9/P9atW4fmzZsbLOv//u//0KVLF1RVVeHatWv4/PPPMXPmTCQmJmLkyJEG5Xv16oUpU6YYjb979+5Gp2uVlJQgIiIC2dnZGDFiBMaOHQsnJydkZ2cjIyMD27Zts0iyfv36daxbtw4eHh4GybqHhwcyMzMNbmaIyP4xWSciyfnggw9w9+5dpKenw8/Pz2C+SqWyQVTy9sgjjyAkJERvWsuWLbF161b8+OOPePbZZ/XmeXt715qs1+Tm5lZv2aNHj2LRokXw8fHB1q1b0aZNG928adOmYfXq1Vi/fj3eeustrFixwuDzgwYN0t1sAEBISAjCwsKwYcMGo8l6hw4dBMdf0xdffIGrV6/ijTfewEsvvWQwX+z6V1JSYnCzVJNCoYCLi4uo6yUieWCfdSKSnKtXr8Ld3d1oog4ASqVS7+9//etfmDFjBp555hn4+/ujf//+mDlzJs6dO2fw2SFDhiA6Ohrnzp1DTEwM+vbtiwEDBmDZsmW4d+8eKioqsHz5cjz11FPo06cPJk2ahEuXLuktIzU1Fb6+vjh27Bjef/99DB48GP7+/hg5ciT2799v0na+9tprePLJJ+Hv748hQ4Zg+fLlKCsr0yv3xx9/ICEhQbeeAQMGICoqCmlpaYLXZUz79u0BAE5OTg1ajhArV67U/ftgoq4VFxeHwMBApKen4/z58/Uuz8/PD61bt8bVq1fFDlW3zAEDBhidX7P+AcDPP/+MKVOmoF+/frquWjt37jQop31+ITs7G//3f/+Hxx57DP369UNqaipefPFFAEBCQgJ8fX3h6+uL6OhoAMb7rD847dtvv8XYsWPRp08fPPnkk1i+fDnu3btnsP5Dhw5h1KhR6NOnD5555hmsW7cOx44dg6+vL1JTU03eV0RkeWxZJyLJ8fT0xJUrV3D48GEMHz683vKffvop3N3dMW7cOCiVSly7dg1ffPEFJkyYgLS0NHh7e+uVv3HjBqZMmYLQ0FCMGDECR48eRVJSEhwdHXHx4kWUl5fjlVdewZ07d7BlyxbMnDkTBw4cgIODfvtGYmIiysrKMGHCBADVSfycOXNQUVGBMWPG1Bnz6dOn8dJLL6Fly5YYP348OnTogHPnziE5ORknTpxAcnIynJyccO/ePUyZMgU3b97ExIkT4e3tjZKSEpw/fx6//PILRo8eLWiflpaW4vbt2wCAsrIynD59Gps3b4aXl5fRLkUVFRW68g9q0qQJWrZsqTetqqrKaFltUp6dnY0zZ84gODgYPXv2NBqfQqFAREQETp06hSNHjtTb77ywsBCFhYVo27at0fn37t0zGpNCoUDr1q3rXLanpyeA6uM5b948NGlS91flN998g9mzZ6Ndu3aYMmUKWrRogf3792PhwoXIyclBfHy8XvnS0lJMnjwZwcHBiIuLw+3bt/Hoo49ixowZWL9+PcaPH49+/foBANq1a1fnuoHqZwi2b9+OqKgojB07Fl9//TW2bNmCVq1aYcaMGbpy//rXvzBnzhx4enpi9uzZcHR0xO7du/HNN9/Uuw4ish0m60QkOX/5y19w7Ngx/PWvf4W3tzeCg4PRp08f9O/fHw899JBB+U2bNhn0cw4PD0dYWBi2bt2Kt956S2/etWvXsGbNGjz33HMAgAkTJmDMmDHYvHkzBg8ejK1bt0KhUAAA3N3dsXjxYhw9ehRPPfWU3nLu3LmDPXv2wM3NTbecUaNGYdmyZQgNDUXTpk1r3cY33ngDSqUSu3bt0usCMWDAAMyePRt79+7FmDFjcPHiRVy5cgXz5s1DbGys8J1oZH01BQcH47333jMa565du7Br1y6D6T179sS+ffv0pl2+fNloK3RmZiZcXFzw3//+FwDQu3fvOmPUzr9w4YLBvJKSEty+fVvXZ33NmjW4f/8+Ro0aZXRZP/zwg9GYmjdvjhMnTtQZR2RkJD799FMkJSVhz549eOSRR+Dv74/g4GAEBwfr3bRVVVXh7bffRvPmzbFz50506NABADBx4kS8+OKL2LBhA0aPHq13w1hQUIAZM2YYJPEDBw7E+vXrERQUZFIXnosXL2Lfvn3o0qULgOp6OHLkSHz66ae6ZP3evXtYunQp2rRpg507d+qeO9DWWSKSLibrRCQ5ffv2xZdffomkpCR8//33SE1N1f1E/8gjj2DZsmXo2rWrrrw2UddoNCgtLYVarUbr1q3RrVs3ZGZmGiy/Q4cOukRdKzg4GGfOnEF0dLQuUdeuDwB+//13g2R9woQJukQdqO67HRUVhVWrVuH48eN4+umnjW7f+fPncf78efz1r3+FWq3WawHu168fmjdvjqNHj2LMmDG65R8/fhxjxoyptSW5PrNmzdJtS1lZGc6ePYutW7di+vTpSEpKMnhodOjQoZg8ebLBcow9/Onh4YF33nnHYLq2e01JSQkA1NsvWzu/uLjYYF5MTIze387OzoiNjUVcXJzRZQUGBhqdJ+QBzVatWiE1NRVbtmzBkSNHcOjQIRw6dAhA9bYuWrQITz75JADgzJkzyM3NRUxMjC5R18b38ssvY9asWfj6668xbdo0vXXU/Lshhg4dqkvUgepfD/r3749PP/0UpaWlcHV1xZkzZ5CXl4eXX35Z71i7uroiKioKiYmJosVDROJisk5EkuTr64tly5YBqB4l4+eff8bOnTvxyy+/YObMmfjyyy/h7OwMADh79izWrl2Ln376yaC/94NJTF3TtAlMzXnaLh8FBQUGnzE2qoi25T8nJ6fWbdP2gX///fdrHTf71q1bAKqTwxkzZmDDhg148skn0atXLzz++OMICQlBQEBAreuoycfHBwMHDtT9PWzYMPTo0QPx8fHYuHEj5s2bp1e+Y8eOeuXr0rx58zrLapNwbdJeG+38B2+AtP7+97+jW7duuHv3Lo4fP47k5GQUFRXV2kWldevWguM3pk2bNpg3bx7mzZuHO3fu4OTJkzhw4AD27NmD2bNnIz09HV5eXrrj3KNHD4NlaLv8ZGdnGyy7ZleihnjwxlXL3d0dQHW9dXV11cVZc4Se2qYRkXQwWSciyfPw8ICHhwfCwsIwceJEZGRkIDMzE4888ghyc3MxadIktGjRAn/5y1/QvXt3NGvWDAqFAkuWLDFI3oG6W1dr9kvX0mg0om2P1tSpUw1a67UeTObi4+MRERGB//f//h9++eUX7Nq1C5s3b8bLL7+M1157zez1a1uH//Of/5i9DCG0SeuZM2fqLHf27FkA1TcWNQUEBOhGgxk6dCjatWuHlStXolevXrpnBiyldevWGDx4MAYPHoxOnTph/fr12L9/P2bOnGnW8po1ayZqfHXVZ0vUWyKyLibrRCQbCoUCgYGByMjIQF5eHgDgyJEjKCsrw0cffWTwoGRBQYGu9d0SLl++bDBN22purPVey8vLC0D1jYHQ1t+uXbsiOjoa0dHRqKiowLRp07Bp0yZMnTrV7K4x2tFCSktLzfq8UF27dkWvXr1w4sQJXLp0yehzBxqNRtdHftiwYfUuc8qUKdi1axfWrFmDkSNH1tvFRiyBgYEAgJs3bwL48zhfvHjRoKx2mrGWb2Me7H4lNg8PDwDAlStXDOYZm0ZE0sGhG4lIco4ePWp02Lny8nIcPXoUwJ/dTbStijVbEL/44guLj8f+2Wef6fWvLi4uxo4dO9CyZUs89thjtX7u4Ycfho+PD3bs2GHQRQKoTqK13W6Ki4tRWVmpN9/FxUXXBaewsNDs+L/66isA9T/4KQbtS5O03Upqev/993Hy5EmEhYXVOmTng5ycnDB9+nQUFBRg27ZtosZ64sQJFBUVGZ2n3Wfabi+9e/dG586dkZqaqlffKisrsXnzZigUinrfCKulfR6gIce0Nv7+/lAqlUhLS9NbfmlpKXbs2CH6+ohIPGxZJyLJWbp0KQoKCjBkyBD4+PigadOmuHHjBvbu3YurV68iPDxcN7TfoEGD0KxZM7z++uuYPHkyWrZsiYyMDHz//ffw9PREVVWVxeJs3bo1IiMjdcM0pqamIjc3F++8806dXR0UCgVWrFiBl156CaNGjcLYsWP13tJ65MgRzJkzB2PGjMHx48fx5ptvYvjw4ejWrRtcXV1x+vRp7Nq1C4GBgfW+jVPrl19+QUVFBYDqm54zZ84gNTVV132opqtXryI9Pd3osgYOHGh0rPG6DBo0CAsXLsTixYsRGhqKMWPGoFu3bro3mJ44cQJPPPGEwcg9dQkLC8MHH3yArVu34sUXX9RrXb9582at8fft21c3PKMxe/fuRWpqKp5++mkEBATA3d0dBQUF+O6773D8+HH06NEDY8eOBVB9s/jmm29i9uzZiIiIwLhx4+Dq6ooDBw7g5MmTmDFjhsHQobXp0aMHXF1dsX37djRt2hQtW7ZEmzZtah3v3RRNmjTB/PnzMW/ePERGRiIiIgKOjo5IS0uDu7s7cnJyLNqyT0TmY7JORJKzYMECfP311/j1119x6NAhFBcXw83NDT4+PoiNjdUbw9zT0xMbN27EqlWrsH79ejg6OiI4OBjJycl4++23cf36dYvFOW/ePPzyyy/Yvn07bt26hW7duiExMdHoGzVr6tWrF9LS0vDxxx/jm2++wY4dO+Dq6goPDw+MHj1al6D5+vri2WefxU8//YS9e/fi/v376NSpE6ZPn46pU6cKjjU5OVn3f0dHR7Rt2xbPPfccZs6cafQBw6NHj+p+xagpKSnJ5GQdgG5s8aSkJOzbtw/5+flo1qwZ/Pz8sGTJEoSHhwsarUWrSZMmeOWVV/CPf/wDW7duxezZs3XzsrKy8Prrrxv93DvvvFNnsh4VFQU3NzccP34cSUlJKCgogJOTE7y8vDB79mxMmTJFb1ScIUOGYOvWrfjoo4+wefNmVFZW4qGHHsI777yDyMhIwdvTtGlTrF69GmvWrMGSJUugVqvx2GOPiZKsA8DIkSPRpEkTfPjhh3jvvffQrl07REREwNfXF7Nnz+YbUokkSqHh0ydERCZJTU1FQkICtm3bhv79+9s6HKIG2bJlC5YvX47PP/8cQUFBtg6HiGpgn3UiIqJGQK1WG3QLKy0tRUpKCtzd3fHwww/bKDIiqgu7wRARETUC2dnZiI2NxfPPP48uXbpApVIhLS0NOTk5eOuttyw6chIRmY/JOhERUSPQpk0bBAUFYe/evcjPz0eTJk3g4+ODuXPnIjQ01NbhEVEt2GediIiIiEii2GediIiIiEiimKwTEREREUkU+6zX4c6dUty/b/1eQm3btkB+fonV10vyxnpD5mC9IXOw3pA5WG+Mc3BQoHVr11rnM1mvw/37Gpsk69p1E5mK9YbMwXpD5mC9IXOw3piO3WCIiIiIiCSKyToRERERkUQxWSciIiIikigm60REREREEsVknYiIiIhIopisExERERFJlKSS9Y0bN8LX1xdhYWEG8zIyMjBhwgQEBgbiiSeewDvvvIO7d+8alFOr1Xj33Xfx5JNPIiAgAOPGjcOPP/5ojfCJiIiIiEQlmWRdpVLho48+QvPmzQ3mZWVlISYmBhUVFViwYAEiIiLw+eefIz4+3qDsggUL8Mknn2DUqFH429/+BgcHB8TGxuLEiRPW2AwiIiIiItFI5qVIK1euhL+/PzQaDYqKivTmrVq1Cu7u7khOToara/Ubnrp06YKFCxfixx9/xIABAwAAmZmZ2L9/PxISEhATEwMACA8PxwsvvIDExESkpKRYdZuIiIiIiBpCEi3rmZmZ2LNnDxISEgzmlZSU4NixYwgPD9cl6gAQFhaG5s2b48CBA7ppBw8ehJOTEyIjI3XTXFxcEBERgV9//RV5eXmW3RAikrzlKRlYnpJh6zDITDx+RNTY2DxZ12g0ePvttxEeHo5evXoZzD9//jzu3bsHf39/venOzs7o1asXsrKydNOysrLQrVs3vaQeAAICAqDRaPTKEhERERFJnc2T9d27d+PixYuIi4szOl+lUgEAlEqlwTylUqnXWq5SqdC+fXuj5QCwZZ2IiIiIZMWmfdZLSkqwcuVKvPLKK0aTbAAoLy8HUN2SXpOLi4tuvrask5OT0XIAUFFRYVJ8bdu2MKm8mJRKN5utm+SL9aZ+Ts6OALivHiSnfcHjJx08BmQO1hvT2TRZ/+ijj+Dk5IQpU6bUWqZp06YAqodkrKmiokI3X1u2srLSaDngz6RdqPz8Ety/rzHpM2JQKt2gUhVbfb0kb6w3wlSqqwCA++p/5FZvePykQW71hqSB9cY4BwdFnQ3ENkvW8/Ly8Mknn+DVV1/FrVu3dNMrKipQWVmJnJwcuLm56bqwaLvDPKhmt5ea3WIeLAeg1tZ7IiIiIjFoH4CePynYxpGQvbBZsp6fn4/KykokJiYiMTHRYP7QoUMRGxuL6dOno0mTJjh9+jSGDx+um69Wq5GVlYWRI0fqpvn5+SE5ORmlpaV6D5meOnVKN5+IGjd+gcobjx8RNTY2S9a7dOmCDz74wGD6mjVrUFZWhjfeeAPe3t5wc3PDgAEDkJ6ejunTp+uS8PT0dJSVlSEkJET32ZCQEGzZsgU7d+7UjbOuVquRmpqK4OBgdOjQwSrbRkREREQkBpsl625ubhg2bJjB9E8++QSOjo568+Lj4xEVFYXo6GhERkbixo0bSEpKwqBBgzBw4EBducDAQISEhCAxMREqlQqenp5IS0tDbm4uli5dapXtIiIiIiISi2TeYFqX3r17IykpCYmJiVi6dClatGiBcePGYc6cOQZlV6xYgTVr1iA9PR2FhYXw9fXFhg0b0K9fPxtETkRERERkPoVGo7H+cCcywdFgSE5Yb8gcrDdkDtab2vEB09qx3hgn2dFgiIiIiOwNk3QSm83fYEpERERERMYxWSciIiIikigm60REREREEsVknYiIiIhIopisExERERFJFJN1IiIiMrA8JUM3DCER2Q6TdSIiIiIiiWKybmfYEkJERERkP5isExERERFJFJN1IiIiIiKJYrJORERERCRRTWwdABEREUnP/EnBtg6BiMCWdSIiIiIiyWLLup1hSwgRERGR/WDLOhERERGRRDFZJyIiIiKSKCbrREREREQSxWSdiIiIiEiimKwTEREREUkUk3UiIiIiIolisk5EREREJFFM1omIiIiIJMpmL0X67bffsH79epw9exb5+flwc3ODn58fZs2aheDgP1/sEx0djZ9++sng86GhoVi9erXeNLVajbVr1yI9PR1FRUXw8/NDfHw8BgwYYPHtISIiIiISm82S9ezsbFRVVSEyMhJKpRLFxcXYu3cvJk+ejI0bN+KJJ57Qle3cuTPi4uL0Pu/h4WGwzAULFuDw4cN48cUX4eXlhbS0NMTGxiI5ORl9+/a1+DYREREREYlJodFoNLYOQuvu3bsYNmwY/P398fHHHwOoblkvKipCenp6nZ/NzMxEZGQkEhISEBMTAwCoqKjACy+8gPbt2yMlJcXkePLzS3D/vvV3j1LpBpWq2OrrJXljvSFzsN6QOVhvyBysN8Y5OCjQtm2L2udbMZZ6NWvWDG3atEFRUZHBvHv37qG0tLTWzx48eBBOTk6IjIzUTXNxcUFERAR+/fVX5OXlWSRmIiIiIiJLsXmyXlJSgtu3b+Py5ctYtWoVLly4YNDH/NKlSwgKCkJwcDCefPJJrF+/Hvfv39crk5WVhW7dusHV1VVvekBAADQaDbKysiy+LSQPy1MysDwlw9ZhEBEREdXLZn3Wtd544w0cOnQIAODk5ISoqCjMmDFDN79r167o378/fH19UVJSgn379mH16tXIzc3FokWLdOVUKhU6dOhgsHylUgkAbFknIiIiItmxebI+a9YsjB8/Hjdu3EB6ejrUajUqKyvh7OwMAFiyZIle+dGjR+PVV1/FF198gZiYGHTv3h0AUF5eDicnJ4Plu7i4AKjuv26quvoPWZpS6Wazdds7J2dHAPa5j+1xm8jyWG/IHKw3ZA7WG9PZPFn39fWFr68vAGDUqFEYO3YsEhIS8N5779X6malTp+LgwYM4fvy4Lllv2rQpKisrDcpqk3Rt0m4KPmBqnyrVVQBgd/uY9YbMwXpD5mC9IXOw3hgnqwdMnZycMHToUBw+fBjl5eW1luvYsSMAoLCwUDdNqVQa7eqiUqkAAO3btxc5WiIiIiIiy5JUsg5Ud2fRaDR1jvySnZ0NAGjTpo1ump+fH65cuWLwuVOnTunmExERERHJic2S9du3bxtMKykpwaFDh9CpUye0bdsWJSUlUKvVemWqqqrw8ccfw8HBQW/UmJCQEFRWVmLnzp26aWq1GqmpqQgODjb68Ck1TvMnBWP+pOD6CxIRERHZmM36rMfFxcHFxQV9+/aFUqnEH3/8gdTUVNy4cQOrVq0CAJw5cwZz587FCy+8AE9PT5SVleHAgQM4ffo0YmNj0bVrV93yAgMDERISgsTERKhUKnh6eiItLQ25ublYunSprTaTiIiIiMhsNkvWR40ahfT0dCQnJ6OoqAhubm4ICgrCihUr8NhjjwEAOnfujODgYBw+fBi3bt2Cg4MDevbsiWXLlmH06NEGy1yxYgXWrFmD9PR0FBYWwtfXFxs2bEC/fv2svXlERERERA2m0Gg01h/uRCY4GgzJCesNmYP1hszBekPmYL0xTlajwRARERFJDd98TbbEZJ3of3gxJiIiIqlhsk5ERDq8aSUikhYm60REREREEsVknYiIiIhIopisExGJhF1IiIhIbDYbZ52IiIhIDvjWa7IlJutE/8OLMREREUkNk3Uiqpe2awdvaOwfjzERkbSwzzoREdk9qT5PINW4iEg6mKwTEVkJEzPL4b4latzs+RrAbjBERCJhFxIiIhIbW9aJiIiIiCSKyToRicqef4okorrx/CcSH7vBEFG92L2DSPo4ahORfWKyTkREdk+qCaxU4yIi6WCyTkRkJUzMLIf7lqhxs+drAPusExERERFJFJN1IiIiImoU5PgQNLvBEJGo7PmnSCKqG89/IvExWSciIrIDTJSJ7BO7wRAAef4sRERERGTvmKwTEREREUkUk3UiImoU+AsiEcmRzfqs//bbb1i/fj3Onj2L/Px8uLm5wc/PD7NmzUJwsH6/u4yMDLz77rs4e/YsWrRogeeeew5z585Fs2bN9Mqp1WqsXbsW6enpKCoqgp+fH+Lj4zFgwABrbhoR2SG+HZKISP7keA23Wct6dnY2qqqqEBkZiTfffBPTpk3D7du3MXnyZBw9elRXLisrCzExMaioqMCCBQsQERGBzz//HPHx8QbLXLBgAT755BOMGjUKf/vb3+Dg4IDY2FicOHHCmptGRERERCQKm7Wsh4aGIjQ0VG/ahAkTMGzYMGzbtg1PPPEEAGDVqlVwd3dHcnIyXF1dAQBdunTBwoUL8eOPP+pazTMzM7F//34kJCQgJiYGABAeHo4XXngBiYmJSElJsd7GERERERGJwKRk/cSJE/j000/x+++/o6CgABqNRm++QqHAV199ZXYwzZo1Q5s2bVBUVAQAKCkpwbFjxzBt2jRdog4AYWFhWLJkCQ4cOKBL1g8ePAgnJydERkbqyrm4uCAiIgKrV69GXl4e2rdvb3Zs9k6OPwtZErs8EJGt8TpkH3gcqaEEJ+u7d+9GQkICmjRpAm9vb3Tq1EmUAEpKSqBWq1FQUIDdu3fjwoULmDVrFgDg/PnzuHfvHvz9/fU+4+zsjF69eiErK0s3LSsrC926ddNL6gEgICAAGo0GWVlZTNaJiIiISFYEJ+sfffQRunXrhqSkJHTo0EG0AN544w0cOnQIAODk5ISoqCjMmDEDAKBSqQAASqXS4HNKpRInT57U/a1SqYzGpf1sXl6eaDETSc3ylAw4OTtiTmSgrUMhkiy2bBKRHAlO1nNzc/H666+LmqgDwKxZszB+/HjcuHED6enpUKvVqKyshLOzM8rLywFUt6TX5OLiopsPAOXl5XBycjJaDgAqKipMjq1t2xYmf0YsSqWbwbSED38AACyd+aS1w2l0nJwdARg/DlIkt3jlKDHuaVuHYDGsN8JY+xos9fNaqnEJZa3jaYnjKOd8QO71xhYEJ+sdO3aEWq0WPQBfX1/4+voCAEaNGoWxY8ciISEB7733Hpo2bQoARtdbUVGhmw8ATZs2RWVlpdFywJ9Juyny80tw/76m/oIiUyrdoFIVG0yvVFcBgNF59ufw6M0AACAASURBVM7aff7ktq8r1VVwcnaUTbwkHbVdb8iQta8LUr4O2UO9sdb+tcR6pFw36mIP9cYSHBwUdTYQCx66MSoqCnv37kVVVZUogRnj5OSEoUOH4vDhwygvL9d1YdF2h3mQSqXS64OuVCqNdnXRfpb91YmIiIhIbmptWf/555/1/vb398fhw4cRGRmJiRMnokuXLnB0dDT43KOPPtqggMrLy6HRaFBaWgofHx80adIEp0+fxvDhw3Vl1Go1srKyMHLkSN00Pz8/JCcno7S0VO8h01OnTunmEwnFvq1EZGu8DtkHHkdqqFqT9ejoaCgUCr1p2qEaFy5caHSeQqHQG6GlLrdv30abNm30ppWUlODQoUPo1KkT2rZtCwAYMGAA0tPTMX36dF0Snp6ejrKyMoSEhOg+GxISgi1btmDnzp26cdbVajVSU1MRHBwsel97IiIiIjng8JHyVmuyvnTpUouuOC4uDi4uLujbty+USiX++OMPpKam4saNG1i1apWuXHx8PKKiohAdHY3IyEjcuHEDSUlJGDRoEAYOHKgrFxgYiJCQECQmJkKlUsHT0xNpaWnIzc21+LYQ2dr8ScHsC0hERGSHFJqabzaykl27diE9PR0XL15EUVER3NzcEBQUhKlTp+Kxxx7TK/vLL78gMTERZ8+eRYsWLRAaGoo5c+agefPmeuUqKiqwZs0a7N27F4WFhfD19cWcOXP0knpTSO0B08aMrQL1Y70hc7DekDlYb+RFKt+hrDfG1feAqeBkPSEhAVFRUQgMND6Oc2ZmJj777DO7asVmsk5ywnpD5mC9IXOw3sgLk3VpE200mLS0NFy7dq3W+Tk5Odi9e7dp0RERERERUa0EJ+v1KSsrQ5MmgodtJyLC8pQMXYsPERERGaozu87NzcX169d1f1++fNlgSEcAKCwsxGeffQYvLy/xIyQiIiIis9m6+ws1TJ3JempqKtatWweFQgGFQoH169dj/fr1BuU0Gg0cHBywZMkSiwVKJHVS6RNIRETUmNj792+dyfqwYcPg4eEBjUaDN954A+PGjUPfvn31yigUCjRv3hx9+vRBp06dLBosERER2X9yQkR/qjNZ9/Pz0735Mzc3F8OHD4ePj49VAiMiIiIiauwEPxE6e/ZsS8ZBREREREQ1CE7WhQzL2LRpU3Tu3BkPP/wwR4YhonrxJ3wiIqK6Cc6oFyxYAIVCoftb+y6lmtMUCgXc3d0RHx+PcePGiRgqEREREVHjIjhZT0pKQmJiIgoLCxEVFYVu3boBqB7O8fPPP0fr1q0xffp0XLt2DSkpKfjHP/6BVq1aYcSIERYLnkhK2EpMRCQePkRLQtl7HRGcrP/6669Qq9XYu3cvmjVrpps+dOhQTJw4EVFRUbhw4QJmzpyJ8ePHIywsDElJSUzWiYiIRGbvyQkR/UnwG0y//PJLjB49Wi9R13J1dcXo0aOxa9cu3d/h4eG4cOGCeJESERERETUygpP1/Px8VFVV1Tr/3r17uHXrlu7v9u3b11meiIiIiIjqJjhZ9/b2xq5du1BSUmIwr7i4GF9++aWuHzsA5OTkoG3btuJESUREhOp+zNq+zEREjYHgPuuzZs1CXFwcQkJCMGbMGHh7ewMArly5grS0NOTn52PNmjUAgPv372P//v0GbzslIiIiIiLhBCfrI0aMwMqVK7F06VJs2LBBb55SqcS7776re5i0qqoKGzduRJs2bcSNloiIiBoFPkRLVM2kNxeFhoZixIgROHPmDHJycgAAHh4e8Pf3h6Ojo66ck5MTunfvLm6kZDUcLouIqPHhtZ9Imkx+zaijoyMCAgIQEBBgiXiIiIiIiOh/TE7WAeDu3bsoKCjQvcX0QZ07d25wUET2gK1URERE1FCCk/X79+9j06ZNSE5O1huisaasrCxRAiMiIqqJN79EVB97aywTnKwnJiZiy5Yt6NmzJ0aMGAF3d3dLxkVERCR59pYUEJH0CE7W9+zZg6eeegobN260ZDxEREREZAO8+ZQmwcl6UVERhg4daslYSCJ4khIRNT689hNJk+Bk3cfHByqVSrQVZ2ZmIi0tDcePH0dubi7c3d3Rt29fxMXFwcvLS1cuOjoaP/30k8HnQ0NDsXr1ar1parUaa9euRXp6OoqKiuDn54f4+HgMGDBAtLiJiIiIiKxFcLI+e/Zs/O1vf0NERAQ6derU4BVv2rQJGRkZCAkJga+vL1QqFVJSUhAeHo5du3bhoYce0pXt3Lkz4uLi9D7v4eFhsMwFCxbg8OHDePHFF+Hl5YW0tDTExsYiOTmZb1Mlq2MrFRERNRbsQmM5gpP106dPo3PnzggNDcWzzz6LLl26wMHBQa+MQqHArFmzBC0vJiYGiYmJcHZ21k0LDQ3FyJEjsXHjRixbtkw3vWXLlggLC6tzeZmZmdi/fz8SEhIQExMDAAgPD8cLL7yAxMREpKSkCNxSIiIi+8WkiuydvdVtwcn6unXrdP/fs2eP0TKmJOvBwYY70tvbGz179sSlS5cM5t27dw8VFRVwdXU1uryDBw/CyckJkZGRumkuLi6IiIjA6tWrkZeXh/bt2wuKjYiISAh7SwqISHoEJ+tff/21JeMAAGg0Gty6dQt+fn560y9duoSgoCBUVlZCqVRi8uTJeOWVV/Ra9rOystCtWzeDZD4gIAAajQZZWVlM1omIiIhqwZtPaRKcrBvrIy62PXv24ObNm4iPj9dN69q1K/r37w9fX1+UlJRg3759WL16NXJzc7Fo0SJdOZVKhQ4dOhgsU6lUAgDy8vIsHj8RERERkZgEJ+sP+v3333Hr1i34+PjAzc1NlEAuXbqERYsWoV+/fnr905csWaJXbvTo0Xj11VfxxRdfICYmBt27dwcAlJeXw8nJyWC5Li4uAICKigqTY2rbtoXJnxGLUinOfqXGhfWGzMF607g4OTsCaPhxt6d6k/DhDwCApTOftHEk8iW0XtlTvbEWk5L1b7/9FosXL8b169cBAFu2bMGAAQOQn5+PqKgozJ07FyEhISYHoVKpMH36dLRq1Qpr1641eHC1pqlTp+LgwYM4fvy4Lllv2rQpKisrDcpqk3Rt0m6K/PwS3L+vMflzDaVUukGlKrb6euWAD0bVjvWGzMF60/hUqqsAoEHH3d7qjRj7xJ4J+e6dExkIoO59aG/1RiwODoo6G4jrzoofcPz4ccyePRutWrXCrFmzoNH8mcS2bdsWnp6e+Ne//mVygMXFxYiNjUVxcTE2bdqk67ZSl44dOwIACgsLddOUSqXRri7aseHZX52IiKg64WKDB5F8CE7WP/jgA/j6+mLnzp2YNGmSwfygoCCcOXPGpJVXVFRgxowZuHr1Kj7++GNdK3l9srOzAQBt2rTRTfPz88OVK1dQWlqqV/bUqVO6+UREUrc8JUPXikVERCQ4Wf/tt98watSoWruodOzYEbdu3RK84qqqKsTFxeHkyZNYu3YtgoKCDMqUlJRArVYbfO7jjz+Gg4OD3ptJQ0JCUFlZiZ07d+qmqdVqpKamIjg42OjDp2S/5JLwyCVOsg3WDyIiEtxnXaPRGH2AU+vOnTt1zq9p2bJl+OabbzB48GAUFBQgPT1dN8/V1RXDhg3DmTNnMHfuXLzwwgvw9PREWVkZDhw4gNOnTyM2NhZdu3bVfSYwMBAhISFITEyESqWCp6cn0tLSkJubi6VLlwqOi4iI5IXPshCRPROcrHfv3h2//vqr0S4wQPXDp6Z0NTl37pzuc99++63ePA8PDwwbNgydO3dGcHAwDh8+jFu3bsHBwQE9e/bEsmXLMHr0aINlrlixAmvWrEF6ejoKCwvh6+uLDRs2oF+/foLjIiIiosaFN3okZYKT9YiICCxevBg7d+7E0KFDAVS/sfTu3btYuXIlTp48ieXLlwtecXJycr1lunbtivfee0/wMl1cXDB//nzMnz9f8GdIXnhBtX9sJSVzXMsrxvKUDNYbIgvgeWVbgpP1iRMnIiMjA2+++SaWL18OhUKBuXPnoqCgAFVVVRgzZgxGjRplyViJiIiIiBoVk8ZZT0xMxIgRI7Bnzx5cvnwZGo0GAQEBCA8Px4gRIywVIxFRo9FYW7Dk8ouKXOIkIvth8htMn332WTz77LMG0y9duoTz588jNDRUlMCIGkIuX6RyiZNsg/WDiIgED91Yn8OHD2Pu3LliLY6IiEiQ+ZOC4dmerzAnIvskWrJORERERETiMrkbDBGRJbHrB5mD9YYaIz5D0TiwZZ2IiIiISKLYsk5kQXJr9ZBbvGQ/5FLn6oqT5w8RWQJb1onI4panZOgSGSIiIhKuzpb1vn37QqFQCFrQvXv3RAmI5I+tS0RERETiqDNZ9/f3t1YcRERENsfGBiKSmjqT9eTkZGvFQUREdbBGEslElUheeK42DuyzTkSSwz7uRERE1TgaDJEFya3VQ27xyh1bsu0LjyMRWQKT9UaOyYJtNLb93li2U060dTAx7mkbR0JEJA57/W5lsk6is7eThIjIkuw1wSAicTBZJyIi+h8mzEQkNUzWiRoptubJiyWO07W8YiR8+APmRAZabB1ERNQwTNaJrICJsWm4n6yL9dP+8RgTyZdJyXpVVRX27t2LH374Afn5+Xjttdfw8MMPo7CwEN9++y0GDBiADh06WCpWqgMvxOLjPiVLY90iIqL6CE7W7969i6lTp+LEiRNo1qwZysvLUVhYCABo0aIFEhMTMXbsWMTHx1ssWBIPE1HbktN+l3JdkXJs1mbqvpg/KZhj2VsZ6ytZAuvVn+x1Hwh+KdL777+P06dPY926dfj666+h0Wh08xwdHTF8+HD88MMPFgmSiIjIXs2fFGy3SQYRNZzgZP3gwYMYP348hg0bBoVCYTDf09MT169fFzU4IiIiMo5v+iVqHAR3g8nLy4Ovr2+t85s1a4bS0lLBK87MzERaWhqOHz+O3NxcuLu7o2/fvoiLi4OXl5de2YyMDLz77rs4e/YsWrRogeeeew5z585Fs2bN9Mqp1WqsXbsW6enpKCoqgp+fH+Lj4zFgwADBcUkNf94iS2GdEoecz9H5k4KhVLph3prvbB0KkWByPueIzCE4WXd3d8fNmzdrnf/f//4X7du3F7ziTZs2ISMjAyEhIfD19YVKpUJKSgrCw8Oxa9cuPPTQQwCArKwsxMTEoEePHliwYAFu3LiBLVu2ICcnB+vXr9db5oIFC3D48GG8+OKL8PLyQlpaGmJjY5GcnIy+ffsKjo1IbPxSISlj/bR/PMZUF0vdAEnhxkoKMTSU4GR9wIABSE1NxbRp0wzmZWdn48svv0RYWJjgFcfExCAxMRHOzs66aaGhoRg5ciQ2btyIZcuWAQBWrVoFd3d3JCcnw9XVFQDQpUsXLFy4ED/++KOu1TwzMxP79+9HQkICYmJiAADh4eF44YUXkJiYiJSUFMGxyZFcKqGcTho5xEhERET2TXCyPnv2bIwdOxYRERF4/vnnoVAo8O9//xvHjh3Djh074OzsjOnTpwtecXCwYSLk7e2Nnj174tKlSwCAkpISHDt2DNOmTdMl6gAQFhaGJUuW4MCBA7pk/eDBg3ByckJkZKSunIuLCyIiIrB69Wrk5eWZ1PJv75iIklBSritSjq02lrphtdW+qG175HRjbi1S2BfmHhceT+niMbF/gh8w9fLywtatW+Ho6Ij33nsPGo0GW7ZswcaNG9GxY0d88skn6NSpU4OC0Wg0uHXrFlq3bg0AOH/+PO7duwd/f3+9cs7OzujVqxeysrJ007KystCtWze9pB4AAgICoNFo9MoSERHJFR8sJWpcTHopkr+/P/bs2YMLFy7g0qVL0Gg08Pb2xsMPPyxKMHv27MHNmzd1Y7WrVCoAgFKpNCirVCpx8uRJ3d8qlcroC5m0n83LyxMlRiIiIilgi2rt7PWXAHvdLqqbScm6lo+PD3x8fEQN5NKlS1i0aBH69eun6/teXl4OAHr92rVcXFx087VlnZycjJYDgIqKCpNjatu2hcmfEYtS6QYASIx72mYxWIKTsyOAP7ePxMX9aijhw+r3Pyyd+aRFlm/KOZrw4Q/IVpWgu0crSR2rhsRS2znNc91yGrJvzf2ssc/Z6tgKOefstf5ZarusuVxrHxN7qAuCk/Uff/wRx44dw9y5c43OX7lyJZ544gk8/vjjJgehUqkwffp0tGrVCmvXroWDQ3XvnKZNmwKoHpKxpoqKCt18bdnKykqj5YA/k3ZT5OeX4P59Tf0FRaZUukGlKrb6eq2hUl0FAHa7fbZkz/WmIaRU5yrVVdBoNKhUV0kiHqDh9aa2/Sul/W5vGrJvzf1szc9J/Xpjr/XPUts1JzJQ9OU++CtAffXGkr8YWGLbxObgoKizgVhwsr5x40a0aFH7gnJycrBx40aTk/Xi4mLExsaiuLgYn332mV6XF+3/td1hHqRSqfQeGFUqlUa7umg/y4dLpcGSP93x50Hb4v4ne8B6TERSIzhZP3fuHF5++eVa5wcGBmLTpk0mrbyiogIzZszA1atXsXXrVnTv3l1vvo+PD5o0aYLTp09j+PDhuulqtRpZWVkYOXKkbpqfnx+Sk5NRWlqq95DpqVOndPOJSH6YPFmfKfu8tjI8XtJk7nHh8SSyHcGjwRQXFxu8MfRBLi4uKCwsFLziqqoqxMXF4eTJk1i7di2CgoIMyri5uWHAgAFIT0/Xeztqeno6ysrKEBISopsWEhKCyspK7Ny5UzdNrVYjNTUVwcHBRh8+JSLb48gWRKaZPymYyTNRIyK4Zb1Dhw44c+ZMrfPPnDljdNSW2ixbtgzffPMNBg8ejIKCAqSnp+vmubq6YtiwYQCA+Ph4REVFITo6GpGRkbhx4waSkpIwaNAgDBw4UPeZwMBAhISEIDExESqVCp6enkhLS0Nubi6WLl0qOC4iIrIc/lJC1mCv9ctet4vqJjhZf+aZZ7Bjxw6EhobqJclA9cOnu3fvRkREhOAVnzt3DgDw7bff4ttvv9Wb5+HhoUvWe/fujaSkJCQmJmLp0qVo0aIFxo0bhzlz5hgsc8WKFVizZg3S09NRWFgIX19fbNiwAf369RMcF5E94ItP/iSlbZk/KZi/IhCRTdnjdd7eCU7WZ8yYgUOHDmHatGkYNGiQrg/4uXPn8P3336Ndu3aYOXOm4BUnJycLLvvII49gx44d9ZZzcXHB/PnzMX/+fMHLJiKqDb/UyN6wTpO1mFLHWB/rJjhZb9euHXbs2IG33noL33//Pb777jsAgEKhwKBBg/Dmm29yxBWyKe3Jzi8j2+D+bjzs+Ryzx20ishR7vhZIiUkvRfLw8MDGjRtRWFiI33//HQDg5eWFVq1aWSQ4Iqnhhcn67G1fy2F75BAjEVFjYdYbTFu1aoWAgACxYyErklPSKadYyXQ8rrbHc4xsjXWQqHYmJ+t3797F9evXUVBQAI3G8O2ejz76qCiBEZHl8QuSrI11jYjINIKT9bt372Lp0qVITU1FVVWVwXyNRgOFQoGsrCxRAyQi00nlxSe8GSAyJLfzorZ4pbAdUoihoay9DXLeV42V4GR98eLF2LVrF55++mk8/vjjcHd3t2RcJFH2cGEkEsqS9ZznEtkC6xuR/AhO1o8cOYLnn38eK1eutGQ8RA3GLyNqzGatrh6p64P4py22Dp5j9ePNGDUGrN/WIThZV6vV6N+/vyVjITKwPCUD1/KK4dnezdahAOBLhoiIpIjXWuG4r+RHcLLu7++Pq1evWjAUsiY5naSe7d1kFW9DWPMiKrUbIbIdezm/GnMSIvdtl2vcRNbgILTg3LlzkZqait9++82S8VAtlqdk8DXlJLrGdCNEjcO1vGJeK0XA7xwi6RDcsv7555+jY8eOGD9+PIKCgtC1a1c4OOjn+gqFAkuWLBE9SCJLkXtrlC2Yss+4X22jQl2F5SkZ3P8SZey4SPlaVFtMUoi15purpay2YyyF/UjSJjhZT0tL0/0/IyMDGRmGJwaTdfvHi4q8STkhaGx4DIQzp96yrhORvRCcrJ87d86ScRCRHZJKwiSVOKzhg/inZdHKaGnX8optuv7GUNdIXPZ8ndJuW2Kc5Uapsmcmv8GUyJqkeNFiNxAisqYHrzmWSOjsIUmsK3Z72D4xcT/Ij8nJellZGU6ePIlbt25h4MCBaNeunSXiImqUrHkRlVu/WaL6PJjMNpTczgW5xElEpjMpWd++fTtWrVqFkpISKBQKbNmyBe3atUN+fj6eeeYZvPnmmxg3bpylYm3UeCEmKZJbQkPUWOusNV6WJURj3f+NCY+x+AQn64cOHcKiRYswdOhQDB48GAsXLtTNa9u2LZ566il89dVXTNZJVqw1nrm11mUN9rId9ozHSH54zOyfvR5je/uOkyLByfrmzZvRv39/fPDBB7hz545esg5UvzRp586dogdItsWT0L7wOJIcmVNv+bIvIvPxu19aBCfrFy5cwLx582qdr1QqkZ+fL0pQZJ6aJ5ccTzY5xqxlbuxy3ub6SGWbpBKHPZF6vZVqXPTni6v4pmZ99lxnOexqwwhO1h0cHHD//v1a5+fl5aFZs2aiBEUkZbyAEMAvFFszd+x1SyVulqwPDy7T0su3R1J9UzOvISSUQ/1Fqvn5+eGHH34wOu/+/fs4ePAg+vTpI1pgRGR98ycF84uDrELqr7OX05sxzSH1/W9vuL+pIQS3rE+ePBlz5szBmjVrEB4eDgDQaDS4fPkyVq9ejYsXL9bZTYZIympr4ah5cZVTItsYWm3ksI1yiNGe1LefpdrKammmjgJjqX0k5vCaJE1SPb/kfC0WnKyHhobi/PnzWL9+PTZs2AAAePnll6HRaKDRaDB79mw8/TTfTCUWOVcq0ifnY8h6SHJhjbrK84Hsgdj1mOeD5Zk0znp8fDyGDx+OvXv34vLly9BoNPDy8kJYWBi7wBDZASYjRPbb9UWIhlwDGsP148HnHqS6nWIcB6luW2Nl8htMe/fujd69e4uy8ry8PGzbtg2nTp3C6dOnUVZWhm3btqF///565YYMGYLr168bfD42Ntag601RURHeffddHDlyBOXl5QgICEBCQgJ69eolSsxSVvPkEnqySekCK4UYzGVu7HLeZrIP5lwDWG/lSYwbkeUpGXBydsScyECzPs/3W1BdeMwMmZysi+nKlSvYuHEjvLy84OvrixMnTtRatnfv3njppZf0pvn4+Oj9ff/+fbzyyiu4cOECpk6ditatW2P79u2Ijo5GamoqPD09LbIdlibWRYcXL+vjPrdfQo6pKaOPsK6Yxpz9ZMl9y+NGRJYiOFlPSEiot4xCocCSJUsEr7x37974z3/+g9atW+Orr77CrFmzai3bsWNHhIWF1bm8gwcP4sSJE/jggw8wbNgwAMBzzz2HESNGYN26dVixYoXg2KhuTCysg/tZPHLfl5aO35pjXwPyOA5yiNFc9rxtUmRv+1vu11O5EZysp6Wl1VvG1GS9RYsWgssCgFqtRlVVVa3juR86dAjt27fH0KFDddPatGmD5557Dvv27UNlZSWcnJxMWic1DrVdcKR6IRJyobRG7LbeP7ZevxC26tvKL1OSA9ZTeZPT8ZPzCyMFJ+vnzp0zmFZVVYXs7Gxs2bIFFy5cwKZNm0QN7kFHjx5FUFAQqqqq0LVrV8TGxmL8+PF6ZbKystC7d28oFAq96X369MHnn3+Oa9eu4aGHHrJYjGKSUyUi+8V6aFly/NKQqsZwcypnrOvSYS/HoDHVqQb1WXd0dIS3tzcWLVqEGTNm4N1338U///lPsWLT8fHxwSOPPAJvb2/cuXMHX3zxBf7+97+jsLAQr7zyiq6cSqXC448/bvD59u3bA6h+oFUuybpUNIaTQCqEXngs+RZGHm+SqvrehSBm3W3M54EYI4jYcxIlh22SQ4xkGtEeMH3qqaewbt06iyTr69ev1/t7zJgxmDhxIj788ENMmDABbm7VSUt5eTmcnZ0NPq+dVl5ebtJ627Y1rZuOmJTKPxMxJ2dHg2k1JXxY/XbZpTOfrLVMbctJjDN9fHwhMZFp+8lY2dqmVairkK0qMVhubesRUj+ElJEzW9TZ+tb54HxTyoqx7gclxj2tO/5SPaednB1x+XohVu08pVdHxTyuUj0HxIzLnOu9qcsQei1L+PAHZKtK0N2jlejHT4ztbKj6jpu5ddda9bSu9Whj1zJlG8S6xpi7/+SYv4iWrBcUFKCsrEysxdXJ0dERL730EuLj43HixAkMGjQIANC0aVOo1WqD8tppTZs2NWk9+fkluH9f0/CATaRUukGlKtb9rR0e68FpWjWH4TJWRshyTFWprhK0LHtuYRHClH1ubJ/WNU2j0ehNr1lv6lu2OWXkzBbbV9vxN3be1ldXTI1faHltvZH68a9UV0Gj0aBSXVXvOdKQdYi1LDFJMS5Trze1TTN2TM1lrf0k9HutrngeXIap8VprO+taj3aeltBY6qo3pjJ3P0jxfHJwUNTZQNzgZL2oqAjHjh3DJ598Itr460J07NgRAFBYWKibplQqkZeXZ1BWO03bHYYaTq7Jt9xuHqw5HrG9s/Uxb2jds3T8dS1fbucN2Z61RxeSMimdP2LFUrPLk5xJ6fjURnCy7ufnZ/DgppZGo0GrVq2wYMEC0QKrT3Z2NoDq0V60/Pz8cOLECWg0Gr1YMzMz0bx5c9mOs05Uk4uzo8l91m355SmHi6FU1fwyFLoPLbWvpfoGR9YxcVlrf0qtHontWl51660lnjGSAjkeOznGLDhZDw8PN5qsu7u7w9vbG88//7zJQzEKUVBQgJYtW8LBwUE3raKiAps3b4arqyuCgoJ000NCQnDo0CF8/fXXunHWb9++jYMHD2Lo0KGyG7ZR6MXS3i8GJC5rfQkzeaqf2PvGmg9hWoNc45Yic2/6GuLBBgIeQ+mT2/kmlzjFIDhZX7ZsmUUC+PDDDwEAly5dAgCkp6fj119/RcuWLTF58mR88803ni/qmQAAIABJREFUWL9+PUaMGAEPDw8UFBQgLS0NV69exVtvvQVXV1fdskaMGIGgoCC8/vrrujeYfvbZZ7h//z7++te/WiR+orqYcvEzpcVUDj89am8iicRQW723p5/j7YX2WPEaQCQO0R4wNdfatWv1/v7yyy8BAB4eHpg8eTJ8fHzQvXt3pKen4/bt23B2dkbv3r2xYMECDB48WO+zjo6O2LBhA1asWIHk5GRUVFSgT58+WL58Oby8vKy2TbYg5KfEmomjJe+ihX5xyu1OviHE2lZzliMkwTc1LiFxVKirbJZEybluPRi7kF/OHtzH1/KKMWv1d3bZvcCW/fZNIaTuWeJGXm5M2S6pvAhOiAdjtcRNy4PbacnrXGMYilMuBCfrP//8s1krePTRR+ucf/78+Trn+/v7GwzdWJdWrVph8eLFWLx4seDPyJnUW1mteXLzgkJywTpKJF9inL/2eg0w93tY7BtceyM4WY+Ojq71AVNjtA95ZmVlmRUYkTnqe4hS6ie5LS5GUt8n9kJK+9nUeial2EkepN6QZE1S+pVLyDNwjW0UHzlsq+BkfcmSJfj0009x5coVjBw5Ej169AAAXLx4Efv27YO3tzcmT55sUkJPZKqaI1HY0522Jd9MCti2b6/Q4yPF42nrmLT1wdT1W+pL98F6ZMkv9frOh5rHRUp1xh7IdX+aM1qRNbqSUPV+dnJ21L1TwpZxAJBVDiE4WS8uLtaNrNKhQwe9eTNnzsT48eNRWlqK6Oho0YNsrKReeUiejNUrS1ywPNu78QGzBtAm26aorY+sXK8lUmqRlDtb7EcpJMtikkuc5hLj1xBe8y3Dof4i1VJSUjB+/HiDRB2ofkHR+PHj8emnn4oaHImHT+b/SdsiaA2Neciy+ZOCOZwoERFZTG3f59b8nrcGwS3rf/zxB5o1a1br/ObNm+OPP/4QJSiyjJqtVJb8+Vro8sUcfcFSNyNi/WQm1rZKJfmX+igWNUdMkOKLfGojpK49ON/YeW1PX1SW0NBWUrHfSGvvrbYNJXS0MymwZp2yRn1hnbQ9wcl6ly5dsGfPHkyYMAEuLi568yoqKrB79254eHiIHiCRUGzFFV9DEggmH0T6xDgnTFmGLYeLtfR6ajai/DenwKw3O4utsV/3zN3/Um/8sTXByfqUKVPw97//HREREZg4cSK6desGALh8+TK2b9+OS5cu4Z///KfFAiV6UG39eeXeWmVOy692GxPjnjZ5fdZsjZLbsbBn1jgGDT3eUui2J9c6K6VWZrnuw4aS63YLjXfW6u8AAB/E1/29I8X9YOzaIsU4HyQ4WR83bhzKysqwdu1a/POf/9SN+qLRaNC0aVO8/vrrGDdunMUCpWpSr1CWVttLGqT05WQuqR5Taw7lZe19IKcXrZjK0i/4sfR+EToKjNB5NctZcuQlshyxX+5m7mfsiZi/+NRl/qRgKJVuUKlsfyOuJZccwqQ3mMbExGDs2LH497//jZycHABA165d8cQTT6Bly5YWCZDq19gvNNRwrEPSw2MiTGNvwCDrYR2rH/eRZZiUrAOAm5sbQkNDLRELWRBPoD9xX9ifxpqw1ewO1ti2vyYpdJ0xRWOtt0RikesLEE1lcrKek5ODH3/8Ebdu3cLIkSPRpUsXqNVq3Lp1C+3atYOzs7Ml4qT/aciXkbW+GKQ8nm9Dl23rCwC/3OtnbB/ZasQPS5o/KRizVn/H7hw1VKirBHXbaujIQA3tOlWzK46UHt40p7ytSeHh0oaSy76uj9zqjhyYlKy/++672Lp1K6qqqqBQKBAUFKRL1p9//nm8+uqriImJsVCoZM+sfXJben32crFqyEsy5L7tRGIT45ywxbB+1jqXxRoG01aMDfMrJfbyvdQYCU7Wd+zYgc2bNyM6OhqDBw/G1KlTdfNatGiBIUOG4Ntvv2Wy3kCWfB2v3H4irovQi01juChZqnVQbI3hWNiSNR8EFkKbsJgbkxS2QwoxmEMqcdtbcmiLd4hIVX2jwGiJPcSoGHXK2C8wUj9egpP17du349lnn8Xf/vY33Llzx2C+r68vfv75Z1GDI0PW+plPahdZY/FIJUZbxWHP3ZrMoU1WTbkpleq2SaVumxKDFOqjZ3s3QcdfattEwm82zRm9w1LHUejL1qRej6z9i4+tySlWLQehBa9evYqBAwfWOr9169ZGk3giIiJLmT8p2KJfvrW908HW7Ol16vMnBcu+v7m9sKd6ZU8EJ+suLi64e/durfNzc3M5fKOdMLV1sjESekGz9wtfQ7ZPzH3TWL/sPdu76Vr1GuP212RsP0jxHBTaIkv1M/X4SrE+ENVHcDeYgIAAHDlyRK+vulZFRQXS09MRHMyLjqU1pE9XY/4yF/q2NbFob3bE3ucN/WKv72VSQl4m82C5htzUNfSG0JR6bu7P0FJ/4FmMl5iY8+Zbocu2Bakmvw8eY23dl8IzI7V1MWysowxJvcuKJYi9zY1p31mL4GR92rRpmDZtGl577TWMHTsWAHDr1i38+9//xvvvv4+bN29i5cqVFguU5MXUPubWPrlt9fZFUzTGLw17wONGZHnWePOkqb+AmHPOW/N6IbVrEn/hEE5wsj5w4EC89dZbWLx4Mfbt2wcAeP311wEATk5OePvtt9G3b1/LRNmICHkdr9RbCYkaE3PPRzm0ZjU0Rl5zGjepJ6+mkmJMliSFh8YbUtaemDTO+vjx4zFkyBAcPHgQly9fhkajgbe3N5577jl06NDBUjGSDdiqL2VtFwcpn6C2ik3M9ZrTJWV5SgYq1FVwcXYULY6GsFWyaonnO0zZFrG/UO395TmmEDpetrVu2Ox9XwOWaW0VY3815GVr9ni8xGTP9VosJr/BVKlUIjo62hKxEAlmbp9KOV0U+JCvNMmpDpHpah5fqR5nKcXFc8J+1DyGUnt/Q2NlcrKude/ePWRmZuLmzZvo0aMHevbsKWZcZEP2elJWqKtMTvAb2tJvr/tSqyGv+G6MD6/JnRzf/CvFc1CqdV+OI4FJpTsob1ikwx6PRZ3J+vHjx3HkyBH85S9/Qdu2bXXTs7OzMWvWLPz3v//VTQsPD8fSpUstFykJJueXMxgjRswfxD9t1S+i2h6stcX+f3Cdtd10PPjTsykPAjeku1RDf/q3p36OYnV7scUbFo2dU1Lf34BlroVCzx1rPBxpTlxaUh9S0tRRrEwh5e22FLlsc231WI55janqTNbT0tJw4sQJLFy4UG96QkICLly4gODgYAQGBuKHH37A7t278dhjj2H06NGCV56Xl4dt27bh1KlTOH36NMrKyrBt2zb079/foOzXX3+NdevW4eLFi2jbti0iIiIwY8YMNGny/9s79/Cqiqv/f5NAAmgkaANaAhUoiQoI4qWiVBSoAhWVeqEKqPUu1kelarDWpxe0CFLUImq9+yJFpKLUUvHaF5+Kl1exoD8QBVRAi0RpIAg5geT8/kjXYe11ZmbP3mefc3bifJ7HR7LP3rPnsmbNmjVrZnuLsH37dtxxxx146aWXUF9fj8MPPxw33XQTDj30UOt8tQbyLbxxOPGlpZMv71vY+HXAtTEQzXGKcanHOHhZ41YnRFzz5YgOldMlm+/INpnKbLY2xUdZt6312FGjsb5y5UoMHjzYc23dunV45513cPTRR2POnDkAgGuuuQZnnHEGnn322UDG+ieffIIHH3wQ3/ve91BVVYX33ntPed/SpUtx1VVX4dhjj8Utt9yCjz76CLNnz8Z//vMf3HLLLan7mpqacNlll+Gjjz7CRRddhE6dOuHPf/4zJkyYgIULF6J79+7WeYszbnBouURl/LRWQyEqr3EuCTsoxK0cKlrbgNcSiavcZ4vWUM4o2iwO7W67wbolEof6DYLRWP/qq6/wve99z3Pt7bffRkFBAc4666zUtXbt2uHUU0/FE088Eejlffr0wZtvvolOnTrh5ZdfxlVXXaW8b/r06TjssMPw8MMPo6io+eSJffbZBw888AAmTJiAgw8+GACwZMkSvPfee5g9ezaGDx8OABg5ciROOeUU3HPPPZg+fXqg/OWLm+79J3Y3NGY0+/14U22gUzpsBDcXwh0k7bgv1fqRzfoMmnYY41h3Pc5KMIp6kROuXHvGZN8OuiEsbLvq0o+rQZ+P86uDfkmTP2tiw5Y6XHXn0hav81oSthuN42zM5uK0oSj6mZNpf4zGekNDA9q1a+e59v777wMAjjnmGM/1gw46CDt27Aj08n333df3nrVr12Lt2rX43e9+lzLUAeC8887D/fffjxdffBGXXXYZAOCFF15A586dMWzYsNR9+++/P0aOHIm//e1v2L17N9q2bRsoj3EjzoZQtogyJrYlDXa5ymdrkKlclkFnnOb6K7lES2m/fOYzH7H8JlryUr3qy7c2dRZ1+2frK9FBje9s5SMu5DJkJt/H0MYZo7F+0EEHeTaRAsC7776LAw44AAcddJDnen19PUpLoxfWVatWAQD69u3rud6lSxcceOCBqd8BYPXq1ejTpw8KCgo89/br1w/z58/Hhg0b0KtXr8jz6NhLnDtHmDzFsRxREOd2ijMt2cjihD2OLdvL4q1RHuPY16L+nkUcy5hrvo1l99OH+Tr2kZxycV71CIrRWD/qqKOwaNEinH322aisrMRLL72Ezz77TBmXvmbNmqx8GKmmpgZA8/nukvLycmzZssVz77HHHpt2X+fOnQE0b2htica6rSK0EcxsdppsGTKZpBeHGGiefj6MvKDlozwGja8nxRwVsn3o7zgaBrn8QBS9QzUI0gCVq0Ey09MZVPfF8auJ+Ugzyr4Up74SBdwQa0n7W4gg+Yt7WXRs2FKHRENj4GdM2IbuRT0WxQGjsX7ZZZfhueeew+mnn46ysjLU1taibdu2uOiiizz3NTY24tVXX8Upp5wSeQbr6+sBAMXFxWm/lZSUYNeuXZ57VffRNUrLlgMO8A/TyRZti4tQXl6a+jcAlJeXev4t2VjTHIbUs2tHtCtpg55dOyrv071Pl67tPW2Li1BQUODJexTY5C3Ms5mkGxbbtuTcdO8/AQBTJ+7d7E3PzlywwvObSmaC5oOWt8fevNg3f5S3GdcOSf2byLS9ZL7aCkOYp29bl5m2uU7GKd2CgoKM5ImHFpjeT//W1ZOpH6p+1+VZlR8bGVZdD3vNr050hOlrQbDNl66MfrqS63PZ5oRKN4TNU9Bn+fNRjB9h8xG1bjfpGRUFBQWoT+zBzAUrUu0QRP8Ssi2D5sPmPTb6xfQu+bxJjgsKCtCUBD7eVJv2my59kvNMdajNWJSP8T8TjMZ6t27dMGfOHMyePRufffYZDj/8cFx55ZVpH0B666230KlTJ0+seFRQzHxDQ0Pab4lEwhNT365dO+V9dE3G3/vx9dc70NSUDPRMVOxuaERNTV3q3wBQU1Pn+bckmUym7k8mk540THCPuOl+07vDvNcWeu/1dzXHBAfxMJjy7FeebDDp7P6pd9q+X3UfpUPeg5qaOpSXlyplRnXOO2HKB8kTvy69PCrZJO9ImHo11Y+qzISqDCqPVKZtrpNxSjeZTGZVnuj99G9VPenyyNue/87lxjYP9D5dfaquZ3ItCEFkPBfoyuinK6mdJ53dP6WjdXIXtFyZ1Ac9y5+3SS/qNshEt5u81bx8qjTks7I/Al49Zpv3IPrOrwym90h4OkHbyCTHVC8yPZO+UY05YZBtqLId/Noo1xQWFhgdxL5fMO3Xrx/uv/9+4z3HHXccnnvuueC5s4DCX2pqalLhLERNTQ2OOOIIz708LIaga/L5loYuzCSbH4gImi59JTSfBD1lIV+fUo765B1+ilAU5VGF7Ni2baKhMW/1aiKKZed8blKO4lxi1aQtSDhL3NpUR7ZD/vze4VfHOjlStW0c6jyMzMUlhCNIWBaNsbZ57t65FB9vqs3ZuBekHcLsSbF5twzt9AtLAvZuwO9VUZbRaXdREBe5DIKvsZ5v6GNGH3zwAfr06ZO6/uWXX2Lz5s2ejx0dcsgheO+995BMJj2bTFeuXIkOHTq0inPWW9JpJlERhYESV+KgNFqDPOW6DKp2611RFvrkg0zkIJ+nqwQhn5sX4ybjcctPEMLmPerjA7NVh0HH2OpxA3HVnUszdlCQsR+HMSFT/PSDzkmWzT1dJsdcS6jzwnxnwI/evXujZ8+emD9/Phob9y5tzJs3D4WFhTj55JNT10aMGIEtW7bglVdeSV3bunUrlixZgmHDhrWIYxunzV2O9Z9v8/wddsY+be5yT6eRf2eDkuKirHW4DVvqjHWRi/JlSlzyGJX3PRN0dcGvZ7pRyO95VR8xeclU9Wbqo3Fpbwkvx033/jPrXsFM6sGv39ukT7/b5CPqNgva12zKmylRO0D8ypiLfhDXvpZNguhHVf1I3cXvsalPv3YvLEBONt1HQdw3pebds37vvfcCaP4yKgAsWrQI7777Lvbbbz+MHz8eAHDjjTfiyiuvxMUXX4xRo0bho48+wty5czF27Fj06NEjldYpp5yCAQMG4MYbb0x9wXTevHloamrC1VdfnfvCRYTtMURRGMmZevgyEfZM3q0LETKl1VqOdiIlIzck8d/DEHYJlZY7JbbtS/mdfd2QrIcERGUU5XrFy+9ducpPpl4q1T3Zznc+Q9+C0NKPB802JjmLe9sG9dxnizAntnBMbdC9c6lSv67/fJv2nZmUleelNa7G591Yv/vuuz1/P/300wCArl27poz1k046Cffccw/uueceTJkyBfvvvz+uvPJKTJw40fNsUVERHnjgAUyfPh1z5sxBIpFAv379MG3atLQvsbZG8q2gdAazww6d4gs74ciFosq3zLVGwkxa47SMm4vY3aBHQ0ZNNsoYZu+R7p44yQPHLya8pY4h3DANW/e6MoeJo4+SsOXJRZ7jKufZIO/G+po1a6zuGz58OIYPH+57X8eOHXHbbbfhtttuyzRrLZagA5SNhznKThG1MmvpZHv5jddbpnGjNm2g86hEiZ8M0e+m/E6b6z0b3S/kTPXOKI0LLgdRfV7eb+MosPeIQN0zQTzRsk4zyWNYeDrkObQNz4u6PR3RkE0ZyUYechEfHdXKcElxkTEvphC/KPpK2PoI8n5eVy11Qph3Y92RTs+uHVPHCnF0whyFd8l22UhnsBBx2AAb5UYmP2Mn6Lts8TNCVfdNnTg4dQxVWIVkE6PoR9D43DAhCdk0hFSGnao+c22MRbXxTlUWHmrHf4s6VCwunrBMQmFsJoLy9yB6JBtGhKneo5Ir0zO2MpRpWIaKTMOybNMP225BV0azHRZjwm+V18Ts65rPaJ+5YAXWbaoNnjmGn2PFjzjYKUFxxrojVrSUeNKoaC1x8982uGGbSRoyPT/i0j9079+wpQ70aYqoDPOW5gEL058znVznWx6ixM/TGyVR1l+mcdKm1a+Whp8+y8eeGtOeqpYwDjtjPWZUjxvo+WhAJgIthZNmtlEiBxlbYyKM0RGXTXUcqez9lH8+B1Ub71ou0SnIMGE7pneYkF70MDJkeibORhRNNHpVlClX8oKcMkFIr6js52HkzHblkMuTKnzLZkCmOvELC4hqUx69L0qZB+xWfaLehGez4dPvXbJeaexSrVLYvtuGXEx66B0fa7zKujyY5C3T8VNODqUcmsLagvYDKp/uIAS/50wOAr/VCZu+T/fEwRmiwhnrDg/5FNKws9vW4l3K1qktmQ7KUcZYRnmfIzw0yGbjWLXCAqSFC8VlAFQZx/kmDh+Siwu2HwzP54bLTAnyUfRshAVxMq2/MKsgPFwzSmwdZi0VZ6zHlEziqfNFHJeS4lZHJlQGRJBTL9oWF3k8pNkYzFpSfUZJtgdNTpgQmzi1Bw3g2Yx1ty1vtjY72xopfpvHEw2NKePN74umUbRxvvuv33sLC4w/Z0wuyh+17NPqX77aLOy3U8JsGA36XJwm3dnGGesxJqxHKhudOsol06DKjIfaqMqWyWAcB68fr9so8yPTytT4mza3+YNBJm+s3wZkXdpB8VPStpvJeN7C5CvKQVT2MZuwtSBhVzx9asPunUvTPF2qz4fbLg8HGdj9+rXpOVk2HbzMNkvlUQ7+cXRemMi3IW8iDnmKKmSzsKDZu57LcSdsXwP0Oi6qvmKKJbfJV9A+pjo8gOcjrjhjvRVhitcM8jxhij2UxP3rXyps49hM123LbOuF03lvgyqUsIZCruP2wr4jyrzZTJC6dy5N+7JptgxMHVGFiZk899lu87gMiFzOgxKll7GkuMjT58PKie2kTVXeqGPXbfOmup6NsKy4y7Su/sOssGVKXPqnH2HyGccQuCA4Y/1bRtxnkXHNlwk/Y882pClMmEVLmSAFVZSZeIJkOkDLlCsVcS8Hb68ojY2WENOtkjVdvqVzI6yXMNuGdhhaUp/T1V9LKkPU5KrsQd+T671TccMZ6y0AW2NPd/SX3+CRiaKXA7JNbDW/r6UrxbChFNksN88Pn0iYDOUg+Ug0NGbV864y8ML2gTDp2OTPJi90j8xX1G0fNnTKbyJkk57faQuq66ZwlDATNBnqw1eG5GkiUYVP+eEXcsPzk42JiG7/i2qMCKL/TSeLRRHOJ/NG7+E6oXvn0rT7dGF7ut+iIsw7eleUYcOWurQ60qXB9UbYFRebe4KspPv9FvR+clapVlZUEyqVHJpWjQA7R1Gc7RBnrH9LkQYdHzRytUxkmmBw/Db3ZZLfTAdK2zLI+4H0fNsuAYcdhLJhuMr0WtqX4cKGloQxRnShYvnwMOVyUOLvMhmMufCe5zp0SaanKntLPdWEiEofBUkjzvXlF04aZMUpzMpJS3eAZYosd9BY+LjijPWYkkk8tQkyejM1zDNRCLmMb8+HwjLFaAY1SFTKWjfI8fP5gxCkLbMRU2rCRlaCylLYM/6vunNpZMfsxWFjsy0tcRIWlKDt4ddnbPVrVJ5822fCTAwy0fWZyk7YcUL33rj2t2+7gQ2EP3VGRbb2fuQTZ6zHmFzE03GvdVRH00Ut8PKDGKqJjCqsxrSszpejpUEY1FsapN5051tTOr0ryjzlU4VR2CLrKVNFSPVy1Z1LU8aIbGtVXcvyENlWjFSnuglG0H5Fm0zlhNf07jDGhgzl8Fum9gu5yHRyIUPcMp2oU/2FWdUI+26/5XyVLORjgrJhSx12JRpx1Z1LMfu6IRkbykHIVYxyvrCZYPE+k82253o9F/WSSeiW7XOZyo8cO2zHwEwdj3E0zFU4Y70VEcRrojIa6d9BDao4CHu+BwITtspEnt3MFblt23KlFjZGNQ7tCUQ3eVTB2yTIpJjOxA56HnQ+vdKt2SOea7KtZ2hCGJRc9FkZV3zTvf/Ebos+KjfPtoTNwjbkckKnc45k0+ERdp+Qbdq2z2UaIkfjCO0VaCkrmhJnrMeMaXPTP27Df7O5ZoIUba4+8JJrohwIbJQV9xKa8AufaC1eAZ0yDLvRLNP2jHJp1RYyalTk2lCJah9KtjcV+xF1GwY1FgDv6pfqd74iQujqK279NkqiWqnIZ/hVS99HkAlh4uTDELSOo2oT3WpG3NvbGeutBF2Yh7xOio97b8ijm4+NGKYOmA9vOe/INh8Aakpm/6t7flB+2xYXpRmD2Rh05GkrQcI9gqzUBA1N0vUB+nf1uIEpQyKI91KWrbDAuyKlWq7lv8vyRG182NSNnETZhiWZJh7yXls5033907QMrkN1KgS/3+bDUkDuJlEUdkdhLn5k+sE3eicQzYkpFKbnR6YGn/SmBt0YHORUoTD9knSI6Uu2su38Npqa3q/7TeVx95MZ2eayrfzC6vj9UfQbnQyawhgpRGza3OVG+Saaki1/RccZ699S+BcMCZOyiHK5jbzMtgMpj5UOg2npK0zcLEcabn5kskEzU2NPZ6hFNSmSdSmNWD9MMpaPjT98EFVNgnTPmI4go3h//nzQZWYZ0x6kPPleAtbFnEY9qcw01jsbA3umeoyTaX/QhXwR0iCL6thXTmsIz5KOCnIIAPlxNgUdI/KhV3MB778U/hLVam2+cMZ6jJHLr7Skn4mwmeJzTUrZz0sQBF1oQpSKw2bADbq5xzZ/KmOKYp117ae6RpOaICcahAlv8iuPzFtUys4mHdXHYqTnLMxmQzrZJSgqr3rYzb9xwlSOXJ8ApMqPrfePYzM5oXu4x9sGv/aLKuxIhaqtbOOLCZXxostz0A828bzYGq6mlWD+btt+E2aPT1BIpwOZ68QoQzxo3IjzPi5ONoxn1eqvDXHTyxxnrLcAEg2NHsWQ6UCgWnrzW9KzSRNAmvEUlQGuCmngf5PClSsFfKmRP6cafIJ62XkdBjXkuYLi3hCeDhkOlOYl015NS48vx/YSsbTyPpv82SDr2GbZVXe+tER1IkMmBrGfEaoK4+AGG99rQArftCmVly2Tuubpq9KyGeC4DFE7yUH8qjuXoqCgAN3K9w2cN1tMsqdbjQl6ylKQfFD6Er7p3iRrUpfwtPwMXworUk2AKH8PVQ+NJGRKNdHJlldRNRFQhWbw9rY5TSlKVGOG6XfTRM+mHlUhKroxJxerXn5jm83ESpXPTHUdtbPUE/LbL7r8mUL7CAqDacmrOc5Yb4VwA8XksbCBGzByk5ku7XwsrQX1/kSNKQa6e+dSrNlYG3jwCaNYVLHSknx5D4JudszEo2eTtt9zciJCRlwmE1rTUZKZ0lqXtFVEYXRK75vN/hPbDct++bPZsBvV5uCokHmdOnFwqO86qFZJ/PYHZUrUfYN0ue1+JZWn28b7rTL0paGcLfyMdz4ByyaZTgb4CUu6UMSWgDPWY0b1OPPHbaTxEqeBOVtLb9IL4heDTs9kis3ATN5Zldeb58l2eV01yEdRlig9arownjCeIelJl6hWIaIMw8n15M7W2IsqX5kMSHIFzsaDFZZct0WQiaDOGcHTkv/Od1ysymGTD3iYkekeWjnmeVWt9OZ7QprNuvTToXwzf77HfZ21NWDaAAAgAElEQVTjIWj7TJu795AGvhKh6z+ZhC0Rug2pqs3qccIZ6zFj2tz0oxtLiouwK9EY2DNLs0k6akwOvjZLtkA4xUiKR7XU6DdT5/dSPnQfEzLx8aZaXHXnUnTvXGr15Unb8vIlaionD1MiTMqXGwsfb6q12q2uegegrpNsK3M+aAQxoEnWqA51cflBvYomJa7yzMh860Jy+FF9qmVtmT7vVzxPpo2nQcpig0qOZZy+KrRKVU9R5YHrA/7RHxMl/z3diPqwCn4qRFg9Bdifb27bNlLOeT2QDNh43/1OG6H2U00Cw6w22vRjSq9tcVHqnHVZJ0EmK6pxLRN9EFQOeFuZQuf49ZLiIt+Po/HQM7/8ht1DozM4+e/T5i73hMMFTVPmk9dBFN5p3SENfh5123ZWhfzqiMtkSIUz1mNMpt4ZnXFnwi8kQGWABvV2hu0IYc6IJ2OwKRlu82VUnhz+7rBpFRY0t6mMl81maAVgPvnEZAjIuOsg75Ftla0wJ5NMkAGXqSHC7w+6pKu7N8wgGWRzupxo2D4DBJfvbLSrKhbedgKka6OoVnXCOB4kNoaHadVK9if6W2Xs2ZJp3egm7rkizOqOn8Hq19b0TrkvzSafOicQjQm2+1ps3hcHgvQ/VblM7csdgnHFGesxhisBHtOXjVmfn9IJ6kENQyZKWnZEnbeCGxRBQ2aC3C+NBb+BubAg3Zjye59KwejqUKeIbAws3cQtk8GVe2ds4QaX7WQxG/JKAyt5fHOxKSzbnh6+NwXw74u2YU9hdYYpnER6ynTpB927kOmeBF3+wkChdap+q9ILNvnm+VFtEJw2dzl2JRoz/m5E2FAgnS7g5dTpm6D9hK+ahZ0s8jFT6iPTe4HsxZqrVgb8HDq5Di266d5/evInT7mj/EaFjcPBds9KPmkRxvpbb72F888/X/nb3//+d/Tq1Sv19/Lly3HHHXdg1apV2HfffTFy5Ej84he/QPv27XOV3ciRijuMYqIOrFJQOkVjI+D8SEl+jQijlFQbcXha5O0kpLKlpfNdib35MH3Ihg9+QTYC8ffKTi5jRk3KkpZU5UAkvWJ86ZU/S8po/efbkEyq3TO60ydkXKtugpCJ9zPoJAfwLudmstE0jPxReIYqNlm3P8EmbzzUSVfftn067H3SqOtdUYa2xUWesCQi0dCYchLwUIGgRliQMDvbZf2wEwCTHOni8nVyoMuzXx64jowiBleHSqdEaZCRvqEwQwCesYWvxqpCnug+lUfZJp+6PqNqX6nndHHLhN/kxy8GH2jWI9JhxN/L5ZhCXcPIta1eVMl/UPmLIixu/efb0uqE58uvDnRtQ2mY0qGQ0zgb5ToK852BIFxwwQWYPn26578uXbqkfl+9ejUuvPBCJBIJTJ48GWeddRbmz5+P6667Lo+5jg4KgfCDGykklB9vsjuNxBRjJ40flWdULvvZxOzZvj9XhMlD74oy9K4o8ygLjmnlIsgSaFRke5VEBZ8YZQOVx11O7DjkTeFxwbr2iwJaPckWtvUbtox+MmPTb7gOyqSvR7miQXnIJD2KC+aOBVN6On0QJhQjCmz1QTb0RmEB0L5Erx+Delo3bKnT9nsuc6a6tgnxs1lVUh33ytOQZTZ9ZEmOt6p+nI3xs3vnUmXbmGTcT8dwnRtE5m3De/zIx7cjMqVFeNaJY445BsOHD9f+PnPmTJSVlWHOnDnYZ599AAAVFRX41a9+hTfeeAODBg3KVVYjR+VVBdTehaAKNewymFxes9koE+ZdQeJmdYSNWZfwwTiMV8xU/kRDozYOkb+b58XkjbH1MAddmqV2jmKC4efh8vMMmyAvCkfWCY/55f3GNHj4lTuT+HqaYNBg5peGX3/SrZRx+dWdz68b0MKGPZngXrFMjfBcL+vnEtuy8fjzqPcEkMFUUODvnrSRFZMjQ4b6Ben/cmUh7P6eoOFGYfH7GrZphZL6lypGXuUMynRS6KevTaGb9Yk9HptBhjeaZJxkxa8f2LS1TMM0YYwDLcpYB4AdO3agXbt2aNOmTdr1ZcuW4eKLL04Z6gBw+umn4/e//z2ef/75FmGsV48zH91oi8rA5TGZpmVemUYmsblBl8ApfEUawbYGEO+k7Uv2dlQ/RUjv5+FCOkNcbsaSYTQqTF8j5QYapaODh1KQYqby9uzaMXU6A6+PsErZdnCjuuDv8ouB97tP53VMNDQqTwfhYRqSsJM03sYc2uhr8zzPj8rbF2QVxvZoMb86lUb71ImDcf1d5qNFpXeTp8H7DOVTtYGRnmlfUpQKETBBoRSyPPLdlD7vf0Fi7k3GjyTTDcJBjGhVv5Phf5kijStVuCDpJpI7kk9+Goxfujw9+l1npEkobME0lvAwFwp/DBLq4DcmypPVZDmDwPWkn7Mn7HGC3APNwyx1k3dbVOWV/a563EClfvDbiPvxptpIJpW6UFypf/i75ClZcaRFhcHccMMNOPLII9G/f39cdNFFWLNmTeq3NWvWYM+ePejbt6/nmeLiYhx66KFYvXp1rrMbimlzl+Ome/8JIPOwARJSLrgfb6rFJdNeNQ5GvFMlGrxHRqryFHQA40t1QZfk6T08T6QMqJM2JZE66pLKE/aYKb5UFwReb/zZsCsEVDbVoBV2SU+2rep9JnQDaBRL+X6yL5d7P95U6/GSU4iLH35yoYr3LSkusjYEZDlMn5EPi6q+ZexmULnj/Uy3DC4pKS5KM/T4e6MO9wqygsjbQcaMm0Kl+DtMMqnqLzrdxutT18+4RzmojqT7+XtUeZfp8pANHSa9oDpmMojshdW1QDThEVRHJpkIqttkn1AZkrTKZ3usK+XV7738/qhWlinNXIRqUWiZ7QoxpymZ7lCxreO40iI8623btsUpp5yCE044AZ06dcKaNWvwyCOP4LzzzsNf/vIX9OjRAzU1NQCA8vLytOfLy8vxr3/9K9fZzhoqJcIF2s+7lGnH5R3VdD64zAcpQcAb0x4FcvDTGQV+Hp+oMQ1uunfqvDgSXWiUCp1yJ2OVD05y6VmFyhMRZMDUeWj4b/JvXTiUX/iVVNA649Y271Eulfr11aCT9Uzl2E/mdKiW37mRG0X/kl566c3LRh8OGi9N+aFnM5EVv/AQQN9XVHKlMvR4X7A5DUfqVjkZiyKMSULe/jB7ADg879mSFy6jKmxCOvmqj0yX/20rXzLcT/U7pZkLz3K7kjboVr6v9ncp00FWIiSy75IMyPpsCUZ8izDWBw4ciIED9wrRsGHDMHToUJx55pm455578Ic//AH19fUAmj3pkpKSktTvQTjgAL1AZYuNNTsAADMXrADQ/ElnAOhVUYZV67/GxpodqRjTmQtW4ONNtWhX0gY9u3YEAJSXl6ZiCdsWF6GgoADtStKbuSnZ/Ht5efNyJj1L/6a0SbnRvTyPvSrKsP7zbQCAn9/1Gnp27eiJf6XfNtbswM/ves3z/oKCglT+6N10vbCgOW0qu6RdSRvUJ/Zg5oIVmDpxcCrPwN7O2aFdc51MnTgYN937z9T76D28zPxvun/959vQs2tHlJeXpspEv9Fz9C6eBpWH1zn9uz6xJ3XfzAUrsLFmR6rdqA0pLQCYce3epc+xNy9O3UfloHcVFBSkygo0H41F9S3LTbQtLvLIDeVx5oIVnvqkNEg21n++LSWbvOw8L5S+fCfljepSXiO5MrURj5OtT+zBxpodaTJP7fTBuq8BNK+y8DRoybVdSRskGhpT7cDTpvzPuHaIp83pt/Wfb0N9Yg/albTx3Ldq/depa7x8G2t2pOqY8gjsbVd6fv3n2zxtxuuroKAA9Yk9qb5Gss/vpzqjvkflqk/s8ZSZBm/KE5c1gvo3lbcp2dyvZlw7JJVv/k5q/531e99F/ZTuob49c8GKlNxSObnumzpxsFYe6D1cBqg/qeRPyhy/h6fH24z6/PrPt6FdSRtP/cj0brr3n6nQC9JbXOfw/sLfy3WLfC/lieoeaNaxuxKN6NCujVZ/lZeXYmPNDiQaGnFYzwM8/YzrdkqT5B9o7idrNtbi53e9hvm3/djzDNUPHw+4rjys5wGp96jqkbdxz64d8fO7XsPO+j0oLPC2lZRxGoMSDY0puaD6pzbhdQ3sDX8h/UbPrFr/tUf+KC+qvsrLLOWExjOuc3kdc9lJNDSm5IfSlu1K75AyzccYyhPpa0qbtwNdr0/sUcp4UxJpuo7XD+l2VZl4u/K/b7r3nx5dIscmXjZ6TtYbzyPvL7xOuBzS+w7reUDaGMP1MN3D3zX25sWoT+zxvJPyy22KONIijHUVhxxyCAYNGoQ333wTANCuXTsAQENDQ9q9iUQi9XsQvv56B5pyfExHMplEQUFBKhaQYtd3/zd2OJlMYp2YJSeTSc/9dHzf7obG1L9pNspn2Os21XpiVWtq6lJfTpUzz90Njam8qNKnv/msWHeMINCskKgciYZGXH/XUlSPG4hu5ftiw5Y6z/sIGeNK9+xmxoesE/qd/03PUpnl35PO7o9pc5d7nqffdv/XeyzrjdIgj4HJe8Lbj+qMv0+mS+Xh5br+rqUp41+2v6pdeFrUtt3K902VhzwOq9Z/neZdrU/sSZWXv4vnkedP9U6eF1k2ukbP1tR4v35LsiHLxfMm37v7vyFDFO8q25nKv2FLXapMdAwdT4d7sGS+KM8y3pvyRH1B1R68PqnM/D6V7NUn9qApqZZt2ZdJNniaTUl4Nncl2G+yDNXjBmLdf0+QojQKC5qfP+eXf0vdd/1dS1MhR4UFzXW6ZmOtp2wkY03J5nhR6l/dyvfFpLP7a8ueTCY9uoHSobzzskqdaOrr/D1As1etW/m+vjIpf+Ppk/dXlkcX003Pc90l+x3VD/9KKeWJ0pXtJvW/Kt+8XIBXJuhvWXfJZBI1NXWpPrW7oRE33fvPVL3ztClP1Ga8rJR3kk+ZT65H1okN4us21ab1Cd4feXtSP6Z+wfcKqOSPj3fJZNIjE7p6U+k4nX6RMk59q6S4yPMbed7pGfle3h5S3uQ7ZX54vdK7TWOkLJOqP8m6oedpHCT9QGUGmidL/B2Tzu6fWhlV5Ze/k8pH90odzWVO1UaqejH1mVxSWFhgdBC3WGMdAA466KCUsU7hLxQOw6mpqUHnzp1zmrdsIZd1VCEtcpnYjzBLXzy+UbWZREeYr5DyEAeb5SqqE758ZhtX77c0ypdPwyydqT4AwTe60SkrpphovnGILwnedO8/PUpIt1wojQPKDz+X3qYcHCkHmW5MlpAM2Nb5hi11aWFCvN1U8ueXV7n0rFtaNoWG2RCkzvhGPRNUZi5/UX0u3K8v87ztSuhj9k15sSmjn47INOxB9bwqX0FCqnTwkDSbc6HD7KkJS2FBs9zJFYMooJANktVMzsPmfYNi+Dl+4RVcN/udcmaD7YZ9AKmN9Kq2V6UjQ2bkRmT+3ZFM947QZDyInPNz+KdOHJxmEKtC2/hzEpuwKD/ivJlURYs21jdu3IhOnToBACorK9GmTRt88MEHOPnkk1P3NDQ0YPXq1Rg9enS+shkJpHioo6iUGBlIEn4SCXVS/sVMP6OKlDOgj2dONDQvoZKHSbWBigwGroBkWvIoPXqvVC4qg4jSJe+PSbn6KRldXXCDTXeyC5D+8ZPeFWVp76S6oEGByk75v2Taq2neNhkPy9tF9VEkypMuHlHG5PJ6lhMLyrNqYOOeKx4jyWVSZ4zZKHy5+Y17c2mpnE7/4UYVfWyEZJxPSmQ5ZCy7jTKnkBJp+Mr4XV1cNe+D9Jyc9FBaUoZIFlVyzttR9TuvA1pCVw2UvJ/xSQ4vK58IcDmg5+kZneHF+xQvny6+lxsb/PQSLq86meKyYjLCqN3Iqy0narLNqJxc/3LZoL/5qSocqmcpQ7LvmeolG8g+oDLyuGxL+QS8/Yru5R8W4xMTVblIvvhET6apyk8U8Ha2+QIo35OlmjxS/VEbylOMJHScIHfucKNblw/+NyH7nyqeO4hcSacR73fytB+pp1QnyASd5MrySccXHTrATw/jhz7w53n+cxW7H5QWcRrM1q1b06698847eOuttzB4cHN8UWlpKQYNGoRFixbhm2++Sd23aNEi7Ny5EyNGjMhZfjOlPrHHV2hVu505dNKH3AHdvqRIe4yh3OVN6avupcEl6AYsnuauRGPqozRB6N65NKV4+IdIKF06trGkuMijPMLidzLJVXcuVX5pkRQEGTHcwA/q3dDtZLfZma8zXky767nxGVSJV48bGGrVgQZkvyP9AK8MAHuP46Kyqk41CmvgBBn8M/EGSpmw2eirgyatlCdbbx4/gYTSCXtqkKwL23zY5I/yFiQ9muBRWkBz29JXNyWmFTSbvNiUV9VXuBGkS1vmgRuzfOVHZdDw9G1PIOGnctA1G4Kc4AGEP7FDp6PJuNVN+qg++D3dO5di9nVDlPk2nYxEqyAm5AlaQVbAJU3J9NBQboTyOlHVa9gVQO5o051kZDMm+em26nED0buiLDWO2+SVG9tB5CionOaDFmGsX3vttbj88stx//33Y/78+bj11lvxs5/9DJ06dcLVV1+duu+6667D1q1bMWHCBMybNw933nknpkyZghNOOAHHHXdcHksQHpXg84GY4MKm6pi7Eo2eWT/3MvJOI2ecfgRVrjLv3MOfrRNZaHYN6BVJpsdkElKpSI8r/13VjkCzF1619Bn2c9R++ZUTJi5L3MAB9q5Y6Lw3EpOhJ1GF1lAegsANc3q3n4dXKmsbefQ7li0oQZemVe/TrUTJ5zJF9hdVSI2tbuCyzSdsMp+8vsmADSJfJnR6webUKt0xcUEHf9tyUD3JiYecGAFeg11lkFAb+a1C6voOn8gGPVqP/s+9nX4rKZkgZZbebSp7EPmSK1B+fZDSluMvfzfhFxZCz8oxQsq1zBelpzJu+bPZHJ9t4bImnS+7Eo1pYZwq/VNSXITeFWWesSVXq1SZ0iKM9eHDh2Pr1q149NFH8bvf/Q4vvPACTj31VPzlL3/Bd7/73dR9ffr0waOPPori4mJMnToVCxYswDnnnIO77747j7kPBoWqcIOIqB43MBUiwZeBgfQvFZo6tZ+nkISXvPBSAUuPGXmndB5FGmRVCoxm96r8mhQEXxqldCgvJs8H957QEiR5jVTLuPSb9J7zTq5S6KolPRrMTIq8KdkcE0htzPNL9etnuJNnSHWusunLo6rJA72XykcrFrwO/DwpdO693yqH9JbLNPhqDuA13ihURL6Py6U02Hk/uOrOpdqzlflkj8oNNH+ohLcl9YOS4iLfI035f4QuNIzy8PGmWm3b+nnJKG8k9/SJd2DvyTEyfxyVMT5t7vLU6hjfI0J5NCENJ2oPLpuUV9OASjLAvcmqCUz1uIGe6yXFRak+6jdgS0OAGzY8TWl0UV5U9abDpLspJI7qnKC+0ZTcG0aomnjysASSBX7GN+97l0x71SPzJF82K0ekl1X6j9q8fUlR6j9el+Q84qFNPBSQp6lqO5PBL0Pc5FhK1wF9G/G+r9Njunh76XyQk32Sd2oPfsyknAiRrqZy8VU4nfyYJs/8eRonVPUoVw94WBd3/PFVd0pHftBPNw5xvfzxplpPuiSTOpnwQ9o+NHZL+ymOtIiY9fPPPx/nn3++1b1HHXUUnnzyySznKHeQ4MjNcVJZkxJWwWOIKU0aVHlssXyvzoBWeQLkQCt/k+l371yair+n3ymelCs5MkyB9Pg9/r72JUUeRcEVvglZHlO4hIxPDtqppQGfibfI9EU/+U5ZF3JZ3S+tIPXIB17Txjjp7ZHvkIMmyQqHnjN5BuXKQFB4DDZ/Lxnw8sxpGmzXbEw3Gjly4FYNXpR3VR2qQmQoDGLDlrq0Nqa86c6XpjhiORBT3ggZMsJDbXQyxL9AKSfDvH9zqB/SlwalLuGGiYyVV9WVDhlXLGPTc7FSMvu6IZ66V3kDTeFqXHZUYwGtqurO0Odtz1cs/Jg6cTDO+eXfcNWdSz2TM9XzPDwBSP8uhgy/TDQ0evKri7f20000xvG9Qbqx0iS/uxKNaX0qiINM9y5+PDJPK8hm/6DILzBTfDeVk++zCIOsC+qztmNWULgtocpHJmWJEy3Cs+7Yi5xdk+dQKgqbeEQbQ4bPhrmBIAcQvjlS5XXIZCmTPDA8zMfU6cmDoLvPz1MXpUJRGf+metCFxnBUHpKPN9WiPrHHN++kvGhwVbULN47ChnbwfQP83Xy1RK5OmCY/JAMm5HK9rAtV7C/fpCmXh1VL0jIOnv9b1lNT0j5MSGUk+8E9qoRsL1NoQlQhJH6hIlyeuW7QTfx5/qShboL6ha4P2+hDmwm4PJ2Kr9zo0pX4rUZxL6vU+ZygepXkmpwjujzoNuiTV3/Dlr0bkzlyIhU2dMK2D9jIsPRyUx34rRL4xT1L/eXXDpQebwMZykTwcYDqUGXY6w5aMBF2PxFfcSfdrgtVVOkmVR6lnpWeb1Ne/cYMiU3/jkonZgNnrMcULsw83lx1DyEHZhl33JREKgyDG9e0JMpj6HRGD/dkcsEnxUfLy9ywl5OCkuIiT3n40jWlofLE0wBDv8mNLuRhk+EWfOChJW3pOZOfq1eFsfBwB1qe43VLv/FJBY+PU0GTEL6ZhsomjVNT/HtJcfNHjmR4gVwi5nVBS5p02g9/VmWscs8xtS+/j+dPNVHicbRrNtamvEfSg8uXdrmhJo0Xet/s64akvHCqAZgGFyo7La3LCSUPM5LwQVE3EPEwBF7nqvSkvPE8SNmkstH7uPdcN7mjuqHlbNkWvSvKUvJGH1pRhbfxf6tkSzXB4V5UbkBT3sn4CGJocjmj9tYt6+vC/LiOIeh5vtTOyyHTVaEKZVAN+DJ8TK6aqvYh8XKojEeVvKvGC1U9qww93UZW/n9i/efbUvU2be5yXDLtVc9qF6XFJ2dUXi7/1G9MhpmMNaZ7TXsF6D1yQihXpflvPPyDwryAZh0idTnXf3J84yFxMn+qcuo2OXPoHWQoy7Kr0qUxk6Na9aBxQLVXSqW/eIgal1+5elNYsHelgOdXTp5IX/DTy1To6pyQ4VfSMUXQu6RT08bJmS+csR4zKJausMB/d7nJK6AbsHRwgwtAatOhbumUe65pQiFReWekF4r/7TdwqwZ3lYIg5CCoCvXx8wBIaICTisfP6+t3LJXq96Cz/PrEHs+pDTwExs8rqTrxgfBTYFF4I6gNHqoeqpysmZBGvSpm3BQPL+FyZnsiCw9DUCENWB1U9mlzl3smN5Suql5sBxceBsDTWf/5trR25/qDJsEq+OSC93c+iIeBl5evaoTZBM4NFnm0rEQVYqiC9wm/kIWgfYMbmyZ43asMLWkskjfZJH9+fZl0XffOpakPG5GxrdKjZACrDGaTzjWFtan2JeiQ+1pkeIacnKo2KqryY9r7EwRqN943bcom82h6zm8/S1jdLXVZUxJpR0vrQmNt9j6o3mdDkJVxLhM0qXCedUfO4UaMLoxADsSmTShA+iYqG/gzurRN4R9Bl7vkM6rYWx3SKxX03X6nymQSYhPWWAkKn4yYYndVG6Rk/mR5VeEmPG1pHOuWWelZndeHZEmuCtFzqhAAbmyGja/0O8tYwutQ905deI5qEkUeLL8wMSDzjVSqybMufILep3snGYCkB8h7SO8wbUazbacgk3Kd0aYyinibr9moXjHhSJnn7emnb1TyrPMy25zOQl5Nk/EUtadRtcIh4WWpHjdQu6ma/u+3L0HV7nKlQrfSRe/w6y98DLMZB+SqE12XKxMyL346hRx9tC+NO6v4s6ajT7mM8skyGbSmECoTiYbGtPApVd3KSZ5qhYLnza8MOk98nD3pEmesx5DDeh6Q5i2Wy2zUAblHgxvnqphx1XXA3ntI7+KhMryT+cWnSWNEetXpP64cTEvltJQP7PWWUVl4OvS3TadUDSCJhsY0LwxXzNJbxZdSVQpeLgFyuncuxUPVQz1eMZ4WKUhupKiWGSV80gZ4BzdVzLMq/MkvDGDNxlpcMu3VtFUCmijqjDuObqMQ4A3VkcqX8kbGkqruKK/cY00hIjyMS1e+MJNGPunh3nVV2VVhVoRuMuvnCSVPER9kZRl21u/xbIjlnl0e9kQnhKiWqbmhrYphlydF6DyT0hPK7yEDga8OUEgV9Tk58eMn1vA6A7y6UHUyCEcu5/P3qFYJgb19jetMKiNHlT/+QRe+2bp63EA8MnkoqrqVpeqS1xN/pzzRg/QYRxVK0buiDFXdyrQrlpQ/3QqElCOuj4N4pE0THzm5oLqSYX9Ut7JvU9/S6TRdGJXJCA8S0rVm495QQNmH+Uq13+SdG/BSj1P/4PKnC98D9q7E6AxYHn6qWrkk5CRX/r3+822e1TjKPwBcdPurKV1Eq8IybM1m7wnJmSyr/IKyKmw4rjhjPeZwQZUbL7jQBllSUgm8HCT9nidMnniV8cMHVJVyk+WgzqrqSDrFqfMS6VYF5CDrp6hkmpQuj9WX7yKjUHpzVPVjEy6hK5cuPtPPqNMNitReQQ1VPpHwe7dOfqjeVM9Lb4+cNPLf+fI2lz1pYHBPHR+MpKefjjQF7PqdKUzCNoSIyyaVgw94PE8q5OCoQ2Voy/z6yYFfuEEUcM8kT5tCMvhRnCqdQv0saL0EzWOY+02heX7nZqvqWuo3Oakh3fRQ9dC0PHNHji5Pfm1r0qv0fND65W1vI1uq9PneD9VvPH88j/KghUxRjcX8QAUTTcn0yV5QaCVLfrhPvsc0bkuC1osqTXqnnByr9mRIB5hJp6pC9FSx/XHDGesxhQwQlRDzWHFdx9HFSHNFY7OErDOIVYqbdxjKu+w0foqFe9d1HmNb48bkxZWxdrZwBaLyZPA243HUwF4jQta56jgz+T6JNFBpQOFxg4RKyZoMLpq4cW+0zIcuRpKMJe4VI2+dSoGrTjKSeZfto9rLYbO0TvlThdkQ3LCWhqlqWVo1YQ4SqiTDYHg+VXmX/1ZNZgYHglsAACAASURBVG2MF+n99EPqIinnVE86g860rE/w7xCQV80W7o0keF9V3R9m8kn5ppAVeqfO4NTtyVGtVOlCvUyEddhwrznVhU1YA+0R0jlEuKc3iq9C0gqBXAnMNIQLCH6KCoUMESonkmk85scvyzT4NW5s8vHXNNnh/1bdp+qXPMyMVj3IaJUrSDxdVbvTXjuOKuy2KYnUXgdT/9ZNVHVhPKa0pFzLTajSuI9rWIwz1mMGFyy5DH7R7a/i4021np3MBFdgchmdlAEpZdrVzg1MYO/ynFyaAtI31Kg2bdLvpKRl55XPU1yqyhOk8jjzUAWVQcQ/HMPzpzIS/GbRqoGG3i+XvVUrEuTxkNdoECNFJg1g8gCTZ5CHO/EwAt62Jk+czDshjQD5LJ/IcWOfe7t1cA+1bkLIPWI6Y4ryLeEyxwcWVTiBnNTI4yS5kWmCD8z8pAcua03JvflVDVTcQOQfayJvPZXHTzZ53fGJoK4MNIHi+SD8POiqrz6q8iGNYmoPmSb/TxooMs82Z03rlvZV+SXkh5yoLCZUezG4wauaVJOOpnqSMcdBjrflcqyTEW64qPQLpSMnKrwueNgYTbopjISnAQDtStooT6yiNGm8AvT7MRINe89T17U35e2qO5em7pErSzzUg+B9k/SqSs+Zzp+nkERedlUIHq1s+cmRDG+jNKmM/IN4hCr+nFCtGgH+3mXy3PMxlYfKEHxskXVI5bBxTvCVP3o+zESO57d3RVnKZkk0NKJ9SVEqPEzlmDIdAuE3fsYBZ6y3MmiJiAueVBCkiPnAqQpn4ct+0uPIDXpubNh4vQHvYGzbSVR5lB+5sEmLOqZuQqEqg854VBnl8ncZ6ydPRwDUCo97IynP3Ggnw18VkiGX+G3aZfZ1Q1KeTT/PFR8gKa8quAFg+6EPKrduk5/MhykdQB8mZIMpzIvCmkxf7yVUx+6p2sVUHmnw2Bj2qnR1Xm5VCIw8hhWAclDXISdLPD9B20Kma5qgmFAZVX71LutAtbIljRzVBmYuz35nrZsw1Z/NyqsKUz9RTfbrE3vS3qF6J02QVZN8W32tyo80mkknqeQtaBjWx5tqcdHtr+KSaa+m3sfzwSfXKuQXr3WonAwc0ylipvAhW+OT67cwBitfkZGTHpWzy4Suvng5+UpApqsr5BDj2IYf5YMW8QXTbxMbttShoKDAdyCz3RTKISEnxUMEVWR8+czmvrDovEIlxUVpH4qQipuMPcC7i12mx+8D0o+ClEuFUSyR2dSJ/GIe3wCnay8bRUtloQ1zKkxxy1SXMs2oUQ3O8mgzaSBxdLJA0LN+KwRrNtb6bqjS5Zvnjx+jySdrPF2+IqGDvtRpMnp0nkKex94VZWnGkwyXMg2IdI9ff6AyyxUmwtSGqnfSxNxP/0Qhk+1L1HKjSp/ywp/xW7VRGa+q+pBfIdbVfaJhr6OGvitgg9QnuvBL6VENCm8zUwiNLt/yvbSCLO/nK5Ck61Q6QPZV1bu5DuBeelkGrlf9HBi2kHxQnnTtwiH9rCpPkL7Gkc/JclN4J0fKtfyb6lSVJoeHm0nDmo/rOl0uVyB5iChvs5aA86y3EPiHUS6Z9mpaXDL3uqkUjgkKK5HCLndOq+ADaBRwhSg9Nvzf8lQL/jsPYfEbtFUxh3TqBI+5pt90AxVfKpSoVhC4p0rnYdNNrrhRR0pSKpw1G9NP/eDGvir+kJ+oEWSFhP6r6pbuYebecVVMthzQaBlTBQ95UT2rgjaX8vfKv4NsLJKn8NDpBRKdt4zqwy8sSYX01lK9q+KWbTZM0Ts7tGuTZgjSv1XhA/x9s68bYjSc5QoaN9b4PgbSayqDpKpbWUou5Ie0eJmB9DAn1e9cZnXtpArboKV3uf9C145kZKjqRNXetOLBl+t7V5R5QhNlOBNPX+eE4XXOQ8Z02BjjupAPafzoZIjni6/80ipVYUFzu/PTbWQe+YolD7OgtqEVX5OeoFNZVO9QobqHnD78XbyNaVKh8uIGMRS5k4KHrvD21e0TkF54MnSp7KbxW9dush+p7uFfO1Wlw2VUJ3cqz79u4shXMPnxrxLTqlQujkcOgzPWY0qYGXBY/Ixy3TK2hBufus2h/F5drLqp7FIZ+tWT9IhydJ5Dlaee59H0tw6Vwcg9Jn5x4DpM3n4ZTy2fk54llZFuCgPh0EAhr9F/ujAfPkjwUCqZX4rNl9dMmOrTr0yqdpXyovMW8+fJkFClxfNgW8d80y8h29kmhInkuynZfHSjrexJI1a12VdnxKnyIFdJVPnwm3jwtFUTY5MOksi2oi9X8rhi00pIUL0tPc1BnB5BvYKqFRe/j7kBXlnm7zZN0OTzprAjXT1mEubAV+FUYYqmEBIVuvL6OXAAeCZbEtO+DZ4ul3+aNJJulm3I+7YKuUJAzwQhKvvEtGnVJNs2OgbwjiV+G4p141RccMZ6zKBYQNXSJKFSbvIeHapTCMLk0QT3eJg6nN/SVRB4XfCzrAF9TLlJ4ejKyJWrX7iBiiCxsvQ+W6+rRBfTq4pZpjS4kletTtgqde5NIa+pRHqKKH2d9yoqbORLbt6m//u1hc4LpYLvPbANLeDeOkIVlhRmRSmIZ1HXt8MuKas8wvRvOXkzTWxNeeIhCoDaWEs0NKbul/tDgPRTYfx0XBh4v5DXTd5JVRp+kMGn0wl+eQqLLh25t8M0VunKaKPrdfuzTJjKHmQ1MgxyQsoPk+DvNYXHmsJzgp4mZUqfyFRnq/YcBU1fFdJpCiPNZB9JtilIJpM59OG2LL7+egeacuniRrNRo5rxqWLu5O/dO5emlrV0Xkqdp5V/tIbHqKniNnVIb4WMgSWqupUpY8hN76ONH7abyqq6lXnu9YuH1GF6jurY1C6y7k3eDm6Y29a/n2yo4G3tV6+q0BVdOeSHj/h9qmdk3ik0QpUXVV1SiICpnEDwGFvdB5xkvnVllZjkPQi6+pSeNV08ddh+bZsnnpcgOoPnUaap+11es30XyYxJDxJUBq5LZL2FzYsKWXdRpk3pyXokOaZ+p3pfFPkw9Q8Or2tb3WbqrzZ6R77XFpUsyHTlPVHqc1WYk+5ZU59S/R1Vf7PZ32YrG0Hh9WFTBr65NB8e9sLCAhxwwL7633OYF4cFUQmtzlDWwcMy+DNBvQU2HgZ5XJrtTD6IMtXFa0cJlTPsUrIKmU+VYZApXHmqQoFslhhVXkW/drepJ4lOydosk8sVlag9oBJd+txDGxV+sixXSQDzknu28hFFGrrVoTDvMU3uAO+RgAQ/NlY1EYpar6gmyKZ7MqUpidRXf1X16ud1jxJ+1K4pxIRDY5dtndika6OrZl83JC1dv42VNitzQfqmlI8wE3A/GQ7S/qojGm3eTwSRa5uxVPXvlooz1mMKN+78OoptR6K0siW4qnABuZHO9JyJoMZFpl91I0yKjMqpy5vKwDSl5adodaErUWBSbLINTcrbdJSY6pqM5Zfn4pvKxz2QfjHiQVDFx0t03msVqnb3yyOfDAUZ/ExeLFnXmaLq10GNBT/9oJOnTMPndH1RtdmewpTCGhU6MknDtIlWh0of8Rh/lS6jCUqmk94g8hYmTDNoDLpETvxt5Fh+IAkIN/EMc4/qPt1qp/RcB51g7ko0Wn2FuLBAvccp2/YGvdsvHMwvH6qvYccJFwZjIB9hMBfd7o3t9VuSt8XGW2ObTqbPc/zCHYhM68F2SS4oQcNzMqm/qEIXJLZLhKolXxkKYkrHNkRCFVoRtt2ikntT2rrlY90zJm+prEtAv7RN9UL3AMFX1Gww9Um/JfcwafuFTwXNZ9D7bNsoSlThNzbv0h0vGdVkLBN55tjqrkxCbvxkKcyzfuGncfbYmsIKo3yHra620RdV3cyhjap3mgg7Zj4yeWjgZzLFhcG0IrK9jG9DkE5vyq9udq9LP46GOuXLtNtfojLGOKY6SzQ0Zuw11hG0XsgTF+Q5Ww+damnXxmiwCd/xSyMItMIQVWxx0JAVfqwnN9KjNtSDeiyDvN9PB9i0O6E6GSNqdHURhW6m/mEjT3SP7cb5bHnDde/KdjtkC7mSJT3KcRiDbQjTb8OUTfY525U9TlB9pVpV0uU9G86tfOGM9ZghhY57WTJZ5stl3CGhG4hz7Y0IY1gGSTssYYxLUzhQh3ZtsjYo686Lln8HCQeJsj38Qoxs0wiC7vzoKN7B5dVULhnGEBSbEDVdunEYCHnd2HzxNmj/kIZAVAaHhIcP2OhpP9nQ3Z8LuFwStrKSyfhg4/zRTSJ09SkdMdnSX1FDbRDkGxJh7AO5LygXdRIkBLE14Yz1mBEmFtEGirvMtcHOMc2A/YjSaynJh7fEZlAKmq/5t/1Y2b428dGtnWwuA+vqN9N6pQl6kI1z2UD3AaA4kA0vLq/HIMez+tWNXz5pghaHSVBrg9rGpm5V4TDZItv9KUhZwh7pHJXeKSxQHxoQZj+G33taIs5Yjxm684+jUuKqDUQqsiHQ2dzcmglxypMpxtomn1J+aINYayeOCjgTuaJns7UiJN9jIlPPmepDQ1ESRL6DrjBWjxtoHfMcZGKVC3KZlziVW0Xc2obIxdhjuxrgFysuicIZwdGt4sk8Z+pwzKfDMhPcBlMD+dhgCqRvMs0mQTeP5YOoYoJbAn4bcE307XUAdv83hv7bQq42eeVrM1kcNrHFIQ8mguYvG5tWAf8N1KYNdkB2NkDHud0cjnxgc9b9Q9Vug6kjBNna4EK7xf2OPMo3vSvKjINO+5Iiz/muNrQvKUJVtzLP3/neGEXeOb6aoronG+/N5PdM08/kHU1J+2dV99mUnb7Gyq9lu86IbBtbUS8x2yD7WCbhcVEQ5btnXzfEuNlN9Vu2PH3ZkJ0o28pmVTffsqEjk/5POiVsuaJ4Lg51mkkdSGQ6JtnyO4YyzJfJc4Ez1lsYUgFXdStTLjPbGJ0lxUXaz7u3ZgoL9n6yWYadRDFJyaYiJONU9Y71n2/z3VAkn4ticuJXXj/lp9o8HWSQzvWmJhnOFfeYVlNd5mIy4DdJysXmd9Ng7rcpOttkO9SJk6k8RRXKaHtgQlxDFvzqwO93Hq5oO14DmbWf7rSpsGQqS/yL6VGjC3sim8d0qlM+vl5qQ6sz1hsaGnDHHXdg8ODBOPzww3HOOefgjTfeyHe2AmGaAfsZMTRzVD1r+w5A3ZltvAm2nkrVvaqJR5CObBOXWNWtLLWaQAYYPSMHkDCeWHmfyViRbdC7okzpmeP3mtqtPrHHN0+qOqLjt+i6zuOhumazudJP+fWuKEvzdpQUF6Wt+oQZ0HhZgrYt1ZfcB9C9c2kqLzqjw8Y74+dZzMSDpjKGw3gDKS3Ts7yObe6nPPHn5NcgVXkw5VX33qpuZajqVuZpo94VZZF6zwoLmt9DctK7ovmdOr0q+5cuL1zG5L9Vk26TPNFz1Kd06fi1m2pPjO6d8neu96lfqbzo1FaFBc1y8VD10JRBm42Vz6D9LOg4qOqP1eMGhhpjSS9m8jVoshH4yjK/Rzc+A/C0Qyb7AHRjrvxdXrO1aQDzYR2qzbRyBTVutDpjffLkyXj88cdx2mmn4eabb0ZhYSEuvfRSvPfee/nOWmi4APGB16Zjc+QAIQVWKkOpcIKGAvjlTdcxVIrI9K7Z1w1Rdr5M86iC5znMYEf3BPEm8sGclHXvijLPoBZ0UqeCDAnVV+hs0pYDUjYxlU0OZvJvLsu8L+lCwvjA2r6kSDn54Onb3KMqR9SySnTvXGpsU91kENgrpzanVFEdcrkk2dUZhyak8aa7R+ZX9Q6TfrCZDHCkLOmMbVVeqC3oWZXBRO8gY1aW369/clTPkw4JcvoHHxt0ExH5Dnq3qt27dy5VGqzciPcrYxT9RfUek1z45YX3AS6/NIGjdgXUZ7hTfakmarb46RmaLAQlTH3LyRpHp1d0bW+SB9N4SnUYRI/EkVZlrK9cuRKLFy/G9ddfjxtvvBFjx47F448/joMOOggzZszId/asaVfSBsBeTwh58mSHI+HjXikbBWcrmA9VD00pZlKiugGX8vpQ9dC0gV/mW2dcy3xySPnpFFf1uIHKNPlHGwoLmu/jg4Rp4OJeKF5vOs8WKQSurLnxQnmguiTvxiOTh+Kh6qFpCpS/k6dBZeDpyed0ckADPdUl5YOXv3rcQKWnkxsf8t/U/vR+XfuGmdyo7lMZQtL44fno3rnUGE8s01UZF9Izo0uLG2PcSKVn+ODBDT0+qeCDvO3ALQc6qg/dwEzpkb6RAx7lmf7j8s/v6965NDWB1PVDXT2ZBljdfeQp95v0qvooXeM6gPKuK58qrSDwZ6NcYpd6STXZJF3DV4jIUKbfdRMv3QSO38PlN4y3l8sKz6NqRYvulys5QeHGMI1HZEibjDp6r0o2qJ8GXbGhcs6+bkjas7qJGvUxlf4ztYFs6+pxA5UyRP2ZVjT88q8bD1W62LQSy3Ueb3uT/NF/Ov3O61COc6p74kybfGcgSpYsWYK2bdvi7LPPTl0rKSnBWWedhTvvvBNbtmxB586d85hDO+bf9mNcf9dSAF7lPm3ucmzYUudRtkRVtzLlb1fduRSJhsbUQErpUNqqNK+6c6knbak46W+ZN+Kh6qGev4G956eSEUjv5r8Be5W3Lk/dO5cqz2KVaarSUMF/p2f5aSrUielrkZR/WYe6fKjeLe/XwRXI7OuGpD0nr22s2YH6xJ5U+1A9UWwgeVR0ciTTl1+m420j80HP0wSgetxAXDLt1bS0uncuTdWv9CpKA5CuUbgSL1P3zqWp+HzdoAak178cEGX7qeSB58/UD0hG6W96l6mOpayoZEMlw9PmLseajbVpg5EOnsddieYv4ZL8tC0uwqSz+3vS/nhTrUdegL0nJJD88Lbm+daVyc9QpU+Vq/QLAFwy7dVU3C3JgkomSWZk/qVMSF3CdSSVk9JONDSm5JX2unB9oKprDj/hi/dBVR3oDFjeJhxZ19QvdAa6zAeVCfDqFJlP3ta9K8rQtrgIu9lXlWX/lP+WDgFeZl0eCVnPsm2I3hVl+HhT84eMZF3SvboxRPZ9nnZhwd664XudVHVMeZM6ydRPVH/rrvGyUrl0fV+WU+ow3qd171OVVSeDAFLyp7I3AODnd72GZDLpSUveM23uco+jQY7rQHp9Ux6pjXV6Vz4X1xh1Sasy1levXo0ePXpgn3328Vw//PDDkUwmsXr16hZhrAN6AfIzPlX3y/Tk8zJNrhjDGpuyk5liUlXPyDxxBRskrTCoZvJyIPd7p+lenTHI30X3+dW1TMs0kaJr2VJOMl052SBoAsmf4QaVTEcawRyaEMjBSldG1XW/9iBsJn6279Jh2zbV4wZ6JrC2afFJjiltaUSo0rGV9yDoDGAgffIjB2CeF+lwUOXRdsBWGXc6z68OW6+zNBilgWxTz7b3yfcSNu0NAFMnDkZNzd56kX1Y1W91xqjOAWOTb9tnuVzxSY0J3URYN6EEgk1Qw/Yb07isMqSlgWubDz5BMek/6SA0pT3/th+jpqbOV8fo0pDyorMTVM/bOsriSKs6Z/3UU09Fly5d8PDDD3uur127Fj/+8Y9x6623erzufuTrnPXy8lKPErQhrrNEm3zlK+9+77VRUtnGtm5UHtJsEhd5U3lk+N+ZpJXr5/NFGH0TJ3JZ79l6Vxx0TVDiKjeqCUGmaeWzTXKdh2y/L1O5iUObZAO/c9ZblbE+fPhwfP/738f999/vub5x40YMHz4ct9xyC8aPH5+n3DkcDofD4XA4HMFoVWEw7dq1w+7du9OuJxIJAM3x60FoSZ51h8PJjSMMTm4cYXBy4wiDkxs136ovmJaXl2PLli1p12tqagCgxcSrOxwOh8PhcDgcQCsz1g855BB88skn+OabbzzXV6xYkfrd4XA4HA6Hw+FoKbQqY33EiBHYvXs3FixYkLrW0NCAhQsXYuDAgejSpUsec+dwOBwOh8PhcASjVcWs9+/fHyNGjMCMGTNQU1OD7t2745lnnsEXX3yBqVOn5jt7DofD4XA4HA5HIFqVsQ4A06dPx1133YVFixZh27ZtqKqqwgMPPIAjjzwy31lzOBwOh8PhcDgC0aqObowadxqMoyXh5MYRBic3jjA4uXGEwcmNmm/VaTAOh8PhcDgcDkdrwhnrDofD4XA4HA5HTHHGusPhcDgcDofDEVNa3QbTKCksLPhWvtvRcnFy4wiDkxtHGJzcOMLg5CYdvzpxG0wdDofD4XA4HI6Y4sJgHA6Hw+FwOByOmOKMdYfD4XA4HA6HI6Y4Y93hcDgcDofD4Ygpzlh3OBwOh8PhcDhiijPWHQ6Hw+FwOByOmOKMdYfD4XA4HA6HI6Y4Y93hcDgcDofD4Ygpzlh3OBwOh8PhcDhiijPWHQ6Hw+FwOByOmOKMdYfD4XA4HA6HI6Y4Yz0mNDQ04I477sDgwYNx+OGH45xzzsEbb7yR72w5ssjKlSvx29/+FqNGjcKAAQNw4okn4rrrrsNnn32Wdu/y5ctx7rnnon///jj++ONx6623YteuXWn3BZEj2zQd8efBBx9EVVUVTj/99LTfnOw4OCtXrsRll12Go48+GkcccQROO+00LFy40HPPK6+8gjFjxqBfv3448cQTcc8992DPnj1paW3fvh233HILjj32WAwYMADnn38+Vq9erXyvbZqO+PHpp5/i2muvxQknnIABAwZg1KhReOCBB9DQ0OC5z+ma7FH0m9/85jf5zoQDuOGGG7Bw4UKcc845GD16NNasWYOHH34YgwYNwkEHHZTv7DmywG233YbXX38dJ510EsaMGYMePXpgyZIlmDNnDn70ox9h//33BwCsXr0a48ePR8eOHXH55Zeje/fueOKJJ7Bq1SqceuqpnjRt5ShImo54U1NTg2uuuQZt27ZFx44dce6556Z+c7Lj4CxduhQXX3wxDjroIJx77rk44YQTUFpaioaGBhxzzDGpe6688kp8//vfxyWXXIKOHTvi4YcfxrZt2zBkyJBUWk1NTbjwwgvx5ptv4oILLsCwYcPw9ttvY86cORgxYgQ6duzoea9Nmo748eWXX2LMmDGora3Feeedh+HDh2PPnj147LHH8Pnnn+Pkk08G4HRN1kk68s6KFSuSlZWVyUcffTR1rb6+Pjl8+PDkeeedl7+MObLKu+++m0wkEp5rn3zySbJv377J6urq1LVLLrkk+cMf/jC5Y8eO1LWnnnoqWVlZmVy2bFnqWhA5sk3TEX+qq6uTEyZMSI4fPz552mmneX5zsuMgtm/fnhw0aFByypQpxvtGjRqVHDNmTHLPnj2pazNnzkwecsghyU8++SR1bfHixcnKysrkSy+9lLr29ddfJ4866qjkDTfcECpNR/z405/+lKysrEx+9NFHnutXX3118rDDDks2NDQkk0mna7KNC4OJAUuWLEHbtm1x9tlnp66VlJTgrLPOwrvvvostW7bkMXeObDFw4EAUFxd7rh188MHo3bs31q1bBwDYsWMHli1bhjPOOAP77LNP6r7TTz8dHTp0wPPPP5+6ZitHQdJ0xJuVK1fir3/9K2666aa035zsODjPPfcctm/fjmuuuQZAc1smk0nPPWvXrsXatWsxduxYFBUVpa6fd955aGpqwosvvpi69sILL6Bz584YNmxY6tr++++PkSNH4uWXX8bu3bsDp+mIH9988w0A4IADDvBc/853voM2bdqgqKjI6Zoc4Iz1GLB69Wr06NHDI5AAcPjhhyOZTGpjAB2tj2Qyia+++gqdOnUCAKxZswZ79uxB3759PfcVFxfj0EMP9ciGrRwFSdMRX5LJJKZMmYIzzjgDhx56aNrvTnYcnDfeeAM9e/bE0qVLMWTIEBx55JE45phjMGPGDDQ2NgIAVq1aBQBp7dulSxcceOCBqd+BZpnp06cPCgoKPPf269cP33zzDTZs2BA4TUf8OProowEAN998Mz788EP8+9//xl//+lc888wzuPTSS1FYWOh0TQ5wxnoMqKmpQefOndOul5eXA4DzrH+L+Otf/4ovv/wSI0eOBNAsG8BeWeCUl5d7ZMNWjoKk6Ygvzz77LNauXYtrr71W+buTHQfns88+w+bNmzF58mSMGTMGs2bNwvDhw/Hggw/i9ttvBxCNzNA1JzOtg8GDB+Oaa67BsmXLcPrpp+PEE0/EDTfcgEsuuQQ///nPAThdkwva5DsDDqC+vh5t27ZNu15SUgIASCQSuc6SIw+sW7cOv/vd73DkkUemTvWor68HgLRwGaBZPuh3utdGjoKk6YgnO3bswB/+8AdcdtllyoEPcLLj8LJz505s27YNv/jFL3DZZZcBAE4++WTs3LkT8+bNw5VXXunbvvwUjvr6euV9dI3SCpKmI55UVFTgmGOOwY9+9COUlZXhf//3fzFr1izsv//+OPfcc52uyQHOWI8B7dq1S8X3cUhoSYgdrZeamhpcfvnl6NixI+6++24UFjYverVr1w4A0o7IAprlg36ne23kKEiajnhy3333oW3btvjZz36mvcfJjoNDbSNP0Rg9ejSWLFmC999/P7DMqO6ja3Svk5mWzeLFi/HrX/8aS5YsQZcuXQA0T/KSySSmT5+OUaNGOV2TA1wYTAzQLenQMpDOc+ZoHdTV1eHSSy9FXV0dHnroIc+yH/2bZIEjlxNt5ShImo74sWXLFjz++OM477zz8NVXX2HTpk3YtGkTEokEdu/ejU2bNmHbtm1OdhweqO2+853veK7T31HJDF1zMtM6+POf/4w+ffqkDHVi6NCh2LlzJz788EOna3KAM9ZjwCGHHIJPPvkkteuaWLFiRep3R+skkUjgiiuuwKeffoo//elP6Nmzp+f3yspKtGnTBh988IHnekNDA1avXu3ZWGgrR0HSdMSPr7/+Grt378aMGTMwbNiw1H8rVqzAunXrMGzYMDz44INOdhwe+vTpA6D53GzO5s2bATSf5ELtJ9v3yy+/xObNm9Nk5v/9v/+X+Nw+3AAADo5JREFUdqLMypUr0aFDB3Tv3h0AAqXpiB9fffVVagMyh7zjjY2NTtfkAGesx4ARI0Zg9+7dWLBgQepaQ0MDFi5ciIEDB6bNaB2tg8bGRlx77bX417/+hbvvvhsDBgxIu6e0tBSDBg3CokWLPMpt0aJF2LlzJ0aMGJG6ZitHQdJ0xI+KigrMnj077b/evXuja9eumD17Ns444wwnOw4P1DZ/+ctfUteSySQWLFiADh06YMCAAejduzd69uyJ+fPnewy0efPmobCwMPUBHEpvy5YteOWVV1LXtm7diiVLlmDYsGGpuOQgaTriR48ePfDBBx+kTvchFi9ejKKiIlRVVTldkwPcF0xjwIEHHoi1a9di7ty5+Oabb7Bp0yZMnToV69atwx133IHvfve7+c6iIwtMnToVzz77LIYMGYJu3bphzZo1qf82bdqU8rL36tULc+bMwdKlS9HU1ISXX34Zd999N44//nhcddVVqfSCyJFtmo74UVJSgp49e6b9R+cO33zzzamv3zrZcRCdO3fGpk2bMHfuXGzevBmbN2/G7Nmz8dprr+Haa6/FscceCwDo2rUrHnvsMSxfvhwNDQ145pln8Oijj2Ls2LEYM2ZMKr2ePXvi9ddfx/z587F79258/PHHmDJlCurq6jBz5kyUlZWl7rVN0xE/unTpgoULF2Lx4sVIJBL46KOPMGvWLPzjH//A2LFjMWrUKABO12SbgqRcw3LkhUQigbvuugvPPfcctm3bhqqqKkyaNAnHHXdcvrPmyBITJkzA22+/rfyta9euePXVV1N/v/POO5gxYwZWrVqFfffdF6NGjcKkSZPQoUMHz3NB5Mg2TUfLYMKECdi+fTsWLVrkue5kx0E0NDTg3nvvxbPPPouvvvoKFRUVuPDCC/HTn/7Uc9/LL7+Me+65B+vWrcP++++PM888ExMnTkSbNt4zKbZt24bp06fj5ZdfRiKRQL9+/TB58uRUyE2YNB3xY+XKlZg1axZWr16N2tpadO3aFWeeeSYuvvhiz4eunK7JHs5YdzgcDofD4XA4YoqLWXc4HA6Hw+FwOGKKM9YdDofD4XA4HI6Y4ox1h8PhcDgcDocjpjhj3eFwOBwOh8PhiCnOWHc4HA6Hw+FwOGKKM9YdDofD4XA4HI6Y4ox1h8PhcDgcDocjpjhj3eFwOL7FTJgwAUOHDs13NvLGrFmzUFVVhU2bNuU7Kw6Hw6HEGesOh8MRkI0bN+KWW27BiBEj0L9/fxx99NEYOXIkqqur8eabb+Y7e7FgwoQJqKqqwrBhw9DQ0JD2OxnJ77//fh5y53A4HC0H951fh8PhCMD777+PCRMmoE2bNjjjjDPw/e9/H/X19fjss8/w+uuvY5999sGxxx6b72zGhk2bNmHevHm44IIL8p0Vh8PhaJE4Y93hcDgCMHv2bOzatQuLFi3CIYcckvZ7TU1NHnIVT9q1a4eKigrcd999OPPMM7HvvvvmO0t5Z8eOHa4eHA5HIFwYjMPhcATg008/RVlZmdJQB4Dy8nLP33//+99xxRVX4MQTT0Tfvn3xgx/8ABMnTsSHH36Y9uzQoUMxYcIEfPjhh7jwwgtxxBFHYNCgQbj99tuxZ88eJBIJTJs2DT/84Q/Rr18/jBs3DuvWrfOksXDhQlRVVWHZsmWYNWsWTjrpJPTt2xejR4/G4sWLA5XzhhtuwODBg9G3b18MHToU06ZNw86dO63TKCwsxKRJk/Cf//wHDz30kO/9pvhxqhtOVVUVJk+ejDfeeANjx45F//79ccIJJ+CBBx4AAGzbtg2//OUvMWjQIPTv3x+XX345vvzyS+W7d+3ahVtvvRXHH388Dj/8cJx99tl44403lPcuW7YMF110EY466ij069cPo0ePxrx587R5XrVqFS6++GIceeSROO2003zrweFwODjOs+5wOBwB6N69Oz755BO8+OKLOPnkk33vf+KJJ1BWVoZzzjkH5eXl2LBhA5566imce+65eOaZZ3DwwQd77t+8eTN+9rOfYdSoUTjllFPw+uuv49FHH0VRURHWrl2L+vp6XHbZZfjPf/6DRx55BBMnTsTzzz+PwkKv72XGjBnYuXMnzj33XADNRvykSZOQSCTwk5/8xJjnDz74ABdccAH2228/jB07Fl26dMGHH36IOXPm4L333sOcOXPQtm1bq/oaNmwYjjzySDz++OMYN25c2mQmU1atWoV//OMfOOecc3D66afj+eefxx/+8AeUlJTg2WefRdeuXfHzn/8cGzZswJw5c1BdXY3HHnssLZ3q6moUFhbi0ksvxY4dOzB//nxccsklePDBB3Hcccel7ps/fz5+/etfY8CAAbjiiivQvn17LFu2DL/5zW+wYcMGVFdXe9L94osvcMEFF2DEiBE4+eSTA012HA6HAwCQdDgcDoc1y5cvT/bp0ydZWVmZPPnkk5OTJ09Ozp07N7l27Vrl/d98803atbVr1yb79OmT/PWvf+25ftJJJyUrKyuTf//73z3Xx4wZk6yqqkpeccUVyaamptT1xx9/PFlZWZl87bXXUteefvrpZGVlZfLEE09Mbt++PXV9+/btyRNPPDF59NFHJ3ft2pW6Pn78+ORJJ53ked/o0aOTp5xySrKurs5z/cUXX0xWVlYmn376aU3t7GX8+PHJAQMGJJPJZPLdd99NVlZWJm+55ZbU73/84x+TlZWVyZUrV6Zd27hxY1p6J510UnL8+PGea5WVlcmqqqrkv/71r9S1RCKRPP7445NVVVXJKVOmeO7//e9/n6ysrEyuW7cu7Z1nnXVWMpFIpK7/+9//Tg4YMCA5YsSI1LUvv/wy2bdv3+SkSZPS8jdlypTkIYccktywYYMnz5WVlcmnnnpKX1EOh8PhgwuDcTgcjgAcccQRePrppzFmzBjU1dVh4cKF+O1vf4tRo0Zh3Lhx2Lhxo+f+Dh06AACSySR27NiBrVu3olOnTujRowdWrlyZln6XLl0wcuRIz7WBAwcimUxiwoQJKCgoSF0/6qijAACfffZZWjrnnnsuSktLU3+Xlpbipz/9KbZt24a33npLW741a9ZgzZo1OPXUU9HQ0ICtW7em/jvyyCPRoUMHvP766xY15c3/8OHD8fTTT+OTTz4J9KwfAwYMQP/+/VN/FxcXo1+/fqn64pjq68ILL0RxcXHq7wMPPBCjR4/G+vXrU6FGL7zwAhoaGnDWWWd56mXr1q0YOnQompqasGzZMk+6ZWVlvisZDofDYcKFwTgcDkdAqqqqcPvttwMAPv/8c/zf//0fFixYgHfeeQcTJ07E008/nTL8Vq1ahbvvvhtvv/12WghERUVFWtqqax07dlT+tt9++wEAamtr057p2bNn2rVevXoBgPFMcTJMZ82ahVmzZinv+eqrr7TP65g0aRL+8Y9/YObMmdp0w9CtW7e0a2Hqi+pGdW3jxo3o1atXqm4uvPBCbX5k3XTr1g1FRUWGEjgcDocZZ6w7HA5HBnTt2hVdu3bF6aefjvPOOw/Lly/HypUrcdRRR+GLL77AuHHjsO++++LKK69Ez5490b59exQUFOD3v/+9Mn7ZZNjJuHQimUxGVh7ioosuwg9/+EPlb2T0BqFXr174yU9+ggULFmDFihXKe/iqgWTPnj3K66b60v0Wtr7ouWnTpqFz587Ke+TkoX379qHe5XA4HIQz1h0OhyMCCgoK0L9/fyxfvhxbtmwBALz00kvYuXMn7rvvvrSz12traz1hF1Gzfv36tGvkGVZ574nvfe97AJonBnxjZRRcffXV+Nvf/oYZM2bgmGOOSfudPOLbtm3z5DGRSKCmpiaVt2ywbt26tBN+qL7IAKfNwJ06dYq8bhwOh0OHi1l3OByOALz++utKL299fX0qlpvCJ8izKz25Tz31VNbPY583bx7q6upSf9fV1eHJJ5/EfvvtpzSUicMOOwyVlZV48skn0+LvgWYPtyqMxIYuXbrg/PPPx9tvv42lS5em/U7GsIz7fuyxx9DU1BTqnbY89thjni+tbt68Gc899xx69OiRas+RI0eiuLgYs2bNQn19fVoadXV1yq+1OhwORyY4z7rD4XAEYOrUqaitrcXQoUNRWVmJdu3apQy7Tz/9FGeccQaqqqoAACeccALat2+PG2+8EePHj8d+++2H5cuX47XXXkP37t3R2NiYtXx26tQJZ599dmpz48KFC/HFF1/g1ltvNYZmFBQUYPr06bjgggtw2mmn4cwzz/R8pfWll17CpEmTQm+avPTSSzF//ny8//77ab8dd9xx6NGjB/74xz+itrYWFRUVePfdd7FixQp06tQp1PtsaWxsxLhx4/DjH/8Y33zzDZ588kkkEgn86le/St1z4IEH4je/+Q1+9atfYdSoUTjttNPQtWtXbN26FR999BFefvllLF682Lhy4XA4HEFxxrrD4XAEYPLkyXjllVfw7rvv4oUXXkBdXR1KS0tRWVmJSy+91GPEdu/eHQ8++CBmzpyJ+++/H0VFRRg4cCDmzJmDKVOm4PPPP89aPq+//nq88847+POf/4yvvvoKPXr0wIwZMzB69GjfZw899FA888wz+NOf/oRXX30VTz75JPbZZx907doVY8aMwaBBg0Lnq7S0FFdeeSWmTp2a9ltRURHuu+8+3HrrrXjiiSfQtm1bHH/88XjiiSdS58Vni2nTpuHJJ5/Egw8+iO3bt6c2ER9//PGe+84880wcfPDBeOSRRzB//nzU1dWhrKwMPXr0wDXXXBP5OfIOh8NRkMzGziSHw+Fw5IWFCxfipptuwv/8z//gBz/4Qb6z43A4HI4McTHrDofD4XA4HA5HTHHGusPhcDgcDofDEVOcse5wOBwOh8PhcMQUF7PucDgcDofD4XDEFOdZdzgcDofD4XA4Yooz1h0Oh8PhcDgcjpjijHWHw+FwOBwOhyOmOGPd4XA4HA6Hw+GIKc5YdzgcDofD4XA4Yooz1h0Oh8PhcDgcjpjy/wHFBXI1xlgQMgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7PaeohUVegmv"
      },
      "source": [
        "# Sort the two lists together by the length of the input sequence.\r\n",
        "train_samples = sorted(zip(full_input_ids, train_labels), key=lambda x: len(x[0]))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_9nyx-BtepE4",
        "outputId": "0dd82ede-cd7c-4877-ccbf-a8d8bf3f234a"
      },
      "source": [
        "train_samples[0:5]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[([101, 13073, 102], 2),\n",
              " ([101, 22678, 37157, 102], 2),\n",
              " ([101, 13073, 30514, 102], 2),\n",
              " ([101, 32470, 91929, 102], 0),\n",
              " ([101, 22678, 37157, 102], 2)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 45
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rBxJhEieetFh",
        "outputId": "9527be0c-cbae-445c-b056-e0c29e70b639"
      },
      "source": [
        "print('Shortest sample:', len(train_samples[0][0]))\r\n",
        "print('Longest sample:', len(train_samples[-1][0]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shortest sample: 3\n",
            "Longest sample: 400\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P44bFOI-e7hW"
      },
      "source": [
        "# Get the new list of lengths after sorting.\r\n",
        "sorted_lengths = [len(s[0]) for s in train_samples]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 427
        },
        "id": "9P5ZX99le_G1",
        "outputId": "e0b2a53f-fa8d-4ffe-ab4a-5195a0ae92c0"
      },
      "source": [
        "\r\n",
        "plt.plot(range(0, len(sorted_lengths)), sorted_lengths)\r\n",
        "\r\n",
        "plt.xlabel('Sample Number')\r\n",
        "plt.ylabel('Sequence Length')\r\n",
        "plt.title('Samples after Sorting')\r\n",
        "\r\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Eh1z0xugfF2t"
      },
      "source": [
        "### Random Batch Selections"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gPrYkFMcgm5H",
        "outputId": "e35abce4-e05f-4932-c122-cac3b2304e7b"
      },
      "source": [
        "train_samples[0]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "([101, 13073, 102], 2)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "isMxwgezfAvO"
      },
      "source": [
        "batch_size = 16"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "s0tjD8pIfJ0S",
        "outputId": "202b67ef-c4b6-4bad-dedf-57fc6d69cfcf"
      },
      "source": [
        "# List of batches that we'll construct.\r\n",
        "batch_ordered_sentences = []\r\n",
        "batch_ordered_labels = []\r\n",
        "\r\n",
        "print('Creating training batches of size {:}'.format(batch_size))\r\n",
        "\r\n",
        "# Loop over all of the input samples...    \r\n",
        "while len(train_samples) > 0:\r\n",
        "    \r\n",
        "    # Report progress.\r\n",
        "    if ((len(batch_ordered_sentences) % 100) == 0):\r\n",
        "        print('  Selected {:,} batches.'.format(len(batch_ordered_sentences)))\r\n",
        "\r\n",
        "    # `to_take` is our actual batch size. It will be `batch_size` until \r\n",
        "    # we get to the last batch, which may be smaller. \r\n",
        "    to_take = min(batch_size, len(train_samples))\r\n",
        "\r\n",
        "    # Pick a random index in the list of remaining samples to start\r\n",
        "    # our batch at.\r\n",
        "    select = random.randint(0, len(train_samples) - to_take)\r\n",
        "\r\n",
        "    # Select a contiguous batch of samples starting at `select`.\r\n",
        "    batch = train_samples[select:(select + to_take)]\r\n",
        "\r\n",
        "    # Each sample is a tuple--split them apart to create a separate list of \r\n",
        "    # sequences and a list of labels for this batch.\r\n",
        "    batch_ordered_sentences.append([s[0] for s in batch])\r\n",
        "    batch_ordered_labels.append([s[1] for s in batch])\r\n",
        "\r\n",
        "    # Remove these samples from the list.\r\n",
        "    del train_samples[select:select + to_take]\r\n",
        "\r\n",
        "print('\\n  DONE - {:,} batches.'.format(len(batch_ordered_sentences)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Creating training batches of size 16\n",
            "  Selected 0 batches.\n",
            "  Selected 100 batches.\n",
            "  Selected 200 batches.\n",
            "  Selected 300 batches.\n",
            "  Selected 400 batches.\n",
            "  Selected 500 batches.\n",
            "\n",
            "  DONE - 536 batches.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rf91Dd3LhBaP",
        "outputId": "ee7d183b-3a23-475a-8b16-c881633ad084"
      },
      "source": [
        "print(batch_ordered_sentences[0])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[101, 1352, 52388, 29400, 66985, 16471, 24458, 1353, 111402, 79960, 18395, 24458, 102], [101, 137, 23080, 10676, 139, 20578, 15190, 10824, 32037, 17094, 10727, 10133, 102], [101, 56326, 40799, 15292, 75301, 13622, 1353, 27921, 34590, 17896, 29400, 119, 102], [101, 24625, 10192, 10114, 67742, 19962, 10731, 13998, 11023, 10107, 10114, 59411, 102], [101, 1328, 43877, 75301, 1334, 59009, 17878, 69203, 1331, 96696, 1323, 37591, 102], [101, 12404, 54047, 11008, 51428, 10174, 33705, 10112, 16642, 10537, 21427, 35397, 102], [101, 13672, 41493, 10107, 119, 31119, 11759, 15688, 12177, 1367, 17676, 21403, 102], [101, 17376, 119, 119, 119, 25460, 11591, 48765, 12078, 119, 119, 119, 102], [101, 158, 10350, 69191, 40300, 18995, 18643, 91189, 10133, 119, 119, 119, 102], [101, 100, 182, 19998, 10271, 13723, 14325, 11534, 87600, 13998, 13061, 11147, 102], [101, 17376, 119, 119, 119, 25460, 11591, 48765, 12078, 119, 119, 119, 102], [101, 109499, 69857, 54047, 19172, 10144, 44082, 61972, 10284, 119, 119, 119, 102], [101, 138, 139, 89696, 21318, 11369, 65729, 11259, 119, 119, 119, 119, 102], [101, 16777, 25230, 10105, 11083, 55352, 41549, 12443, 15914, 10853, 59731, 13660, 102], [101, 137, 77603, 23315, 77603, 23315, 182, 11565, 39041, 19314, 16200, 22659, 102], [101, 85581, 11147, 15694, 55170, 13723, 10799, 30771, 10116, 18028, 21533, 10112, 102]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qmkQK_gVgiTu",
        "outputId": "7fdef324-a5db-41de-c5f5-a711fa43c216"
      },
      "source": [
        "batch_ordered_labels[0]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0, 1, 1, 1, 1, 1, 1, 0, 1, 1, 0, 1, 2, 1, 1, 1]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 53
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xZ0TFA8Mfn6m"
      },
      "source": [
        "### Padding"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c9rdswcpfTip"
      },
      "source": [
        "py_inputs = []\r\n",
        "py_attn_masks = []\r\n",
        "py_labels = []\r\n",
        "\r\n",
        "# For each batch...\r\n",
        "for (batch_inputs, batch_labels) in zip(batch_ordered_sentences, batch_ordered_labels):\r\n",
        "\r\n",
        "    # New version of the batch, this time with padded sequences and now with\r\n",
        "    # attention masks defined.\r\n",
        "    batch_padded_inputs = []\r\n",
        "    batch_attn_masks = []\r\n",
        "    \r\n",
        "    # First, find the longest sample in the batch. \r\n",
        "    # Note that the sequences do currently include the special tokens!\r\n",
        "    max_size = max([len(sen) for sen in batch_inputs])\r\n",
        "\r\n",
        "    #print('Max size:', max_size)\r\n",
        "\r\n",
        "    # For each input in this batch...\r\n",
        "    for sen in batch_inputs:\r\n",
        "        \r\n",
        "        # How many pad tokens do we need to add?\r\n",
        "        num_pads = max_size - len(sen)\r\n",
        "\r\n",
        "        # Add `num_pads` padding tokens to the end of the sequence.\r\n",
        "        padded_input = sen + [tokenizer.pad_token_id]*num_pads\r\n",
        "\r\n",
        "        # Define the attention mask--it's just a `1` for every real token\r\n",
        "        # and a `0` for every padding token.\r\n",
        "        attn_mask = [1] * len(sen) + [0] * num_pads\r\n",
        "\r\n",
        "        # Add the padded results to the batch.\r\n",
        "        batch_padded_inputs.append(padded_input)\r\n",
        "        batch_attn_masks.append(attn_mask)\r\n",
        "\r\n",
        "    # Our batch has been padded, so we need to save this updated batch.\r\n",
        "    # We also need the inputs to be PyTorch tensors, so we'll do that here.\r\n",
        "    py_inputs.append(torch.tensor(batch_padded_inputs))\r\n",
        "    py_attn_masks.append(torch.tensor(batch_attn_masks))\r\n",
        "    py_labels.append(torch.tensor(batch_labels))\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "roS5lLWdX-5J"
      },
      "source": [
        "### Check the number of token reductions because of smart batching"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wlh7cteVfroV",
        "outputId": "81873144-ca65-43b6-c3cd-ff8b0cf551b1"
      },
      "source": [
        "# Get the new list of lengths after sorting.\r\n",
        "\r\n",
        "padded_lengths = []\r\n",
        "\r\n",
        "# For each batch...\r\n",
        "for batch in py_inputs:\r\n",
        "    \r\n",
        "    # For each sample...\r\n",
        "    for s in batch:\r\n",
        "    \r\n",
        "        # Record its length.\r\n",
        "        padded_lengths.append(len(s))\r\n",
        "\r\n",
        "# Sum up the lengths to the get the total number of tokens after smart batching.\r\n",
        "smart_token_count = np.sum(padded_lengths)\r\n",
        "\r\n",
        "# To get the total number of tokens in the dataset using fixed padding, it's\r\n",
        "# as simple as the number of samples times our `max_len` parameter (that we\r\n",
        "# would pad everything to).\r\n",
        "fixed_token_count = len(train_sentences) * max_input_length\r\n",
        "\r\n",
        "# Calculate the percentage reduction.\r\n",
        "prcnt_reduced = (fixed_token_count - smart_token_count) / float(fixed_token_count) \r\n",
        "\r\n",
        "print('Total tokens:')\r\n",
        "print('   Fixed Padding: {:,}'.format(fixed_token_count))\r\n",
        "print('  Smart Batching: {:,}  ({:.1%} less)'.format(smart_token_count, prcnt_reduced))\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Total tokens:\n",
            "   Fixed Padding: 3,425,600\n",
            "  Smart Batching: 342,700  (90.0% less)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8wgdhsI-YIAL"
      },
      "source": [
        "## Load the model configuration from the transformers library using AutoConfig"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9er7EHvdfudf",
        "outputId": "8f5d0eaf-d873-4cae-c3c5-309ac2ac41dc"
      },
      "source": [
        "# Load the Config object, with an output configured for classification.\r\n",
        "config = AutoConfig.from_pretrained(pretrained_model_name_or_path=model_name,\r\n",
        "                                    num_labels=len(class_names))\r\n",
        "\r\n",
        "print('Config type:', str(type(config)), '\\n')\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Config type: <class 'transformers.models.bert.configuration_bert.BertConfig'> \n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bWvc9Ikv4HtD"
      },
      "source": [
        "# model = AutoModelForSequenceClassification.from_pretrained(pretrained_model_name_or_path=model_name,config = config)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xouf5M2B55ei"
      },
      "source": [
        "# model.cuda()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qYV6GYYUxr4M"
      },
      "source": [
        "## Load the model from the transformers library using AutoModel"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Jmcj0PrXxr4O",
        "outputId": "4979f53d-40f0-411e-fc74-70b5ce23f27f"
      },
      "source": [
        "# Load the pre-trained model for classification, passing in the `config` from\r\n",
        "# above.\r\n",
        "model = AutoModel.from_pretrained(\r\n",
        "    pretrained_model_name_or_path=model_name,\r\n",
        "    config=config)\r\n",
        "\r\n",
        "print('\\nModel type:', str(type(model)))\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Model type: <class 'transformers.models.bert.modeling_bert.BertModel'>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2LEx-3Spxr4R"
      },
      "source": [
        "# freeze all the parameters\r\n",
        "for param in model.parameters():\r\n",
        "    param.requires_grad = False"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bOXKJrU_xr4S"
      },
      "source": [
        "## Creating a custom BERT model "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "m3CbIF1Axr4U"
      },
      "source": [
        "\r\n",
        "class Custom_Model(nn.Module):\r\n",
        "    ## defining the constructor for the class\r\n",
        "    def __init__(self, model,num_labels):\r\n",
        "      ## calling the super class constructor\r\n",
        "      super(Custom_Model, self).__init__()\r\n",
        "      ## adding the bert layer to the model\r\n",
        "      self.model = model        \r\n",
        "      # relu activation function\r\n",
        "      self.relu =  nn.ReLU()\r\n",
        "      # adding a dense layer to our custom model\r\n",
        "      self.fc1 = nn.Linear(768,512)\r\n",
        "      # adding another dense layer to our custom model ,i.e., the Output layer\r\n",
        "      self.fc2 = nn.Linear(512,num_labels)\r\n",
        "      # adding a softmax  activation function for our custom model's output layer\r\n",
        "      self.softmax = nn.LogSoftmax(dim=1)\r\n",
        "\r\n",
        "    #defining the forward pass\r\n",
        "    def forward(self, input_id, mask):\r\n",
        "      #pass the inputs to the model  \r\n",
        "      outputs = self.model(input_id,mask)\r\n",
        "  \r\n",
        "      last_hidden_state = outputs.last_hidden_state       ## last hidden state from the model\r\n",
        "      pooler_output = outputs.pooler_output               ## pooler output from the model\r\n",
        "      ## adding a fully connected layer to the BERT model\r\n",
        "      x = self.fc1(pooler_output)\r\n",
        "      ## applying relu activation function\r\n",
        "      x = self.relu(x)\r\n",
        "      # the final output layer\r\n",
        "      x = self.fc2(x)\r\n",
        "      # apply softmax activation to our output layer\r\n",
        "      x = self.softmax(x)\r\n",
        "\r\n",
        "      return x"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Lqp1b5hNxr4V",
        "outputId": "07b12436-b798-4edd-cfb6-442330a2d061"
      },
      "source": [
        "\r\n",
        "\r\n",
        "print('\\nLoading model ...')\r\n",
        "# pass the pre-trained BERT to our define architecture\r\n",
        "model = Custom_Model(model,num_labels=len(class_names))\r\n",
        "\r\n",
        "model.cuda()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Loading model ...\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Custom_Model(\n",
              "  (model): BertModel(\n",
              "    (embeddings): BertEmbeddings(\n",
              "      (word_embeddings): Embedding(119547, 768, padding_idx=0)\n",
              "      (position_embeddings): Embedding(512, 768)\n",
              "      (token_type_embeddings): Embedding(2, 768)\n",
              "      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "      (dropout): Dropout(p=0.1, inplace=False)\n",
              "    )\n",
              "    (encoder): BertEncoder(\n",
              "      (layer): ModuleList(\n",
              "        (0): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (1): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (2): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (3): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (4): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (5): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (6): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (7): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (8): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (9): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (10): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (11): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "      )\n",
              "    )\n",
              "    (pooler): BertPooler(\n",
              "      (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "      (activation): Tanh()\n",
              "    )\n",
              "  )\n",
              "  (relu): ReLU()\n",
              "  (fc1): Linear(in_features=768, out_features=512, bias=True)\n",
              "  (fc2): Linear(in_features=512, out_features=3, bias=True)\n",
              "  (softmax): LogSoftmax(dim=1)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 62
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LjQSUqCWJTbI"
      },
      "source": [
        "## Custom Loss function"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iOHVl1RRJSUB"
      },
      "source": [
        "# # convert class weights to tensor\r\n",
        "# weights= torch.tensor(class_wts,dtype=torch.float)\r\n",
        "# weights = weights.to(device)\r\n",
        "\r\n",
        "# # loss function\r\n",
        "# criterion  = nn.NLLLoss(weight=weights) \r\n",
        "criterion = SelfAdjDiceLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "evYV0yrQ2u_b"
      },
      "source": [
        "## Loading Optimizer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8tAijXBEh7t8"
      },
      "source": [
        "# Note: AdamW is a class from the huggingface library (as opposed to pytorch) \r\n",
        "# I believe the 'W' stands for 'Weight Decay fix\"\r\n",
        "optimizer = AdamW(model.parameters(),\r\n",
        "                  lr = 2e-5, # This is the value Michael used.\r\n",
        "                  eps = 1e-8 # args.adam_epsilon  - default is 1e-8.\r\n",
        "                )\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s9n3_5jI2ywo"
      },
      "source": [
        "## Loading lr scheduler"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "agidP6D6iBFI"
      },
      "source": [
        "# Number of training epochs. I chose to train for 1 simply because the training\r\n",
        "# time is long. More epochs may improve the model's accuracy.\r\n",
        "epochs = 4\r\n",
        "\r\n",
        "# Total number of training steps is [number of batches] x [number of epochs]. \r\n",
        "# Note that it's the number of *batches*, not *samples*!\r\n",
        "total_steps = len(py_inputs) * epochs\r\n",
        "\r\n",
        "# Create the learning rate scheduler.\r\n",
        "scheduler = get_linear_schedule_with_warmup(optimizer, \r\n",
        "                                            num_warmup_steps = 0, # Default value in run_glue.py\r\n",
        "                                            num_training_steps = total_steps)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MDDp05Cm245e"
      },
      "source": [
        "## Training Loop"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "H4ERR8fBiEJn",
        "outputId": "9d4cd80a-4dae-4ae7-dfa0-a0ccfce11707"
      },
      "source": [
        "# We'll store a number of quantities such as training and validation loss, \r\n",
        "# validation accuracy, and timings.\r\n",
        "training_stats = {\r\n",
        "    'epoch':[],\r\n",
        "    'train_loss':[],\r\n",
        "    'Training Time':[],\r\n",
        "    'val_loss':[],\r\n",
        "    'Validation Time':[],\r\n",
        "    'train_acc':[],\r\n",
        "    'val_acc':[]\r\n",
        "}\r\n",
        "\r\n",
        "# Update every `update_interval` batches.\r\n",
        "update_interval = good_update_interval(total_iters=len(py_inputs), num_desired_updates=10)\r\n",
        "\r\n",
        "# Measure the total training time for the whole run.\r\n",
        "total_t0 = time.time()\r\n",
        "best_accuracy = 0\r\n",
        "# For each epoch...\r\n",
        "for epoch_i in range(0, epochs):\r\n",
        "\r\n",
        "    predictions = []\r\n",
        "    true_labels = []\r\n",
        "    \r\n",
        "    # ========================================\r\n",
        "    #               Training\r\n",
        "    # ========================================\r\n",
        "    \r\n",
        "    # Perform one full pass over the training set.\r\n",
        "\r\n",
        "    print(\"\")\r\n",
        "    print('======== Epoch {:} / {:} ========'.format(epoch_i + 1, epochs))\r\n",
        "    \r\n",
        "    # At the start of each epoch (except for the first) we need to re-randomize\r\n",
        "    # our training data.\r\n",
        "    if epoch_i > 0:\r\n",
        "        # Use our `make_smart_batches` function (from 6.1.) to re-shuffle the \r\n",
        "        # dataset into new batches.\r\n",
        "        (py_inputs, py_attn_masks, py_labels) = make_smart_batches(train_sentences, train_labels, batch_size,tokenizer,max_input_length)\r\n",
        "    \r\n",
        "    print('Training on {:,} batches...'.format(len(py_inputs)))\r\n",
        "\r\n",
        "    # Measure how long the training epoch takes.\r\n",
        "    t0 = time.time()\r\n",
        "\r\n",
        "    # Reset the total loss for this epoch.\r\n",
        "    total_train_loss = 0\r\n",
        "\r\n",
        "    # Put the model into training mode. Don't be mislead--the call to \r\n",
        "    # `train` just changes the *mode*, it doesn't *perform* the training.\r\n",
        "    # `dropout` and `batchnorm` layers behave differently during training\r\n",
        "    # vs. test (source: https://stackoverflow.com/questions/51433378/what-does-model-train-do-in-pytorch)\r\n",
        "    model.train()\r\n",
        "\r\n",
        "    # For each batch of training data...\r\n",
        "    for step in range(0, len(py_inputs)):\r\n",
        "\r\n",
        "        # Progress update every, e.g., 100 batches.\r\n",
        "        if step % update_interval == 0 and not step == 0:\r\n",
        "            # Calculate elapsed time in minutes.\r\n",
        "            elapsed = format_time(time.time() - t0)\r\n",
        "            \r\n",
        "            # Calculate the time remaining based on our progress.\r\n",
        "            steps_per_sec = (time.time() - t0) / step\r\n",
        "            remaining_sec = steps_per_sec * (len(py_inputs) - step)\r\n",
        "            remaining = format_time(remaining_sec)\r\n",
        "\r\n",
        "            # Report progress.\r\n",
        "            print('  Batch {:>7,}  of  {:>7,}.    Elapsed: {:}.  Remaining: {:}'.format(step, len(py_inputs), elapsed, remaining))\r\n",
        "\r\n",
        "        # Copy the current training batch to the GPU using the `to` method.\r\n",
        "        b_input_ids = py_inputs[step].to(device)\r\n",
        "        b_input_mask = py_attn_masks[step].to(device)\r\n",
        "        b_labels = py_labels[step].to(device)\r\n",
        "\r\n",
        "        # Always clear any previously calculated gradients before performing a\r\n",
        "        # backward pass.\r\n",
        "        model.zero_grad()        \r\n",
        "\r\n",
        "        # Perform a forward pass (evaluate the model on this training batch).\r\n",
        "        # The call returns the loss (because we provided labels) and the \r\n",
        "        # \"logits\"--the model outputs prior to activation.\r\n",
        "        logits = model(b_input_ids, b_input_mask)\r\n",
        "\r\n",
        "        # Accumulate the training loss over all of the batches so that we can\r\n",
        "        # calculate the average loss at the end. `loss` is a Tensor containing a\r\n",
        "        # single value; the `.item()` function just returns the Python value \r\n",
        "        # from the tensor.\r\n",
        "        loss = criterion(logits,b_labels)\r\n",
        "        # Move logits and labels to CPU\r\n",
        "        logits = logits.detach().cpu().numpy()\r\n",
        "        label_ids = b_labels.to('cpu').numpy()\r\n",
        "      \r\n",
        "        # Store predictions and true labels\r\n",
        "        predictions.append(logits)\r\n",
        "        true_labels.append(label_ids)\r\n",
        "        \r\n",
        "        # Accumulate the training loss over all of the batches so that we can\r\n",
        "        # calculate the average loss at the end. `loss` is a Tensor containing a\r\n",
        "        # single value; the `.item()` function just returns the Python value \r\n",
        "        # from the tensor.\r\n",
        "        total_train_loss += loss.item()\r\n",
        "\r\n",
        "        # Perform a backward pass to calculate the gradients.\r\n",
        "        loss.backward()\r\n",
        "\r\n",
        "        # Clip the norm of the gradients to 1.0.\r\n",
        "        # This is to help prevent the \"exploding gradients\" problem.\r\n",
        "        torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)\r\n",
        "\r\n",
        "        # Update parameters and take a step using the computed gradient.\r\n",
        "        # The optimizer dictates the \"update rule\"--how the parameters are\r\n",
        "        # modified based on their gradients, the learning rate, etc.\r\n",
        "        optimizer.step()\r\n",
        "\r\n",
        "        # Update the learning rate.\r\n",
        "        scheduler.step()\r\n",
        "\r\n",
        "    # Calculate the average loss over all of the batches.\r\n",
        "    avg_train_loss = total_train_loss / len(py_inputs)     \r\n",
        "       \r\n",
        "    training_accuracy = check_accuracy(predictions,true_labels)       \r\n",
        "    \r\n",
        "    # Measure how long this epoch took.\r\n",
        "    training_time = format_time(time.time() - t0)\r\n",
        "\r\n",
        "    print(\"\")\r\n",
        "    print(\"  Average training loss: {0:.2f}\".format(avg_train_loss))\r\n",
        "    print(\"  Training Accuracy: {0:.2f}\".format(training_accuracy))\r\n",
        "    print(\"  Training epoch took: {:}\".format(training_time))\r\n",
        "        \r\n",
        "    (py_inputs, py_attn_masks, py_labels) = make_smart_batches(val_sentences, val_labels, batch_size ,tokenizer,max_input_length)\r\n",
        "    val_loss,val_accuracy,validation_time = eval_model(model,py_inputs, py_attn_masks, py_labels,criterion)\r\n",
        "\r\n",
        "    if val_accuracy > best_accuracy:\r\n",
        "        torch.save(model.state_dict(), 'best_model_state.bin')\r\n",
        "        best_accuracy = val_accuracy\r\n",
        "\r\n",
        "\r\n",
        "    # Record all statistics from this epoch.\r\n",
        "\r\n",
        "    print(\"\")\r\n",
        "    print(\"  Average validation loss: {0:.2f}\".format(val_loss))\r\n",
        "    print(\"  Validation Accuracy: {0:.2f}\".format(val_accuracy))\r\n",
        "    print(\"  Validation epoch took: {:}\".format(validation_time))\r\n",
        "    \r\n",
        "    \r\n",
        "    training_stats['epoch'].append(epoch_i + 1)\r\n",
        "    training_stats['train_loss'].append(avg_train_loss)\r\n",
        "    training_stats['Training Time'].append(training_time)\r\n",
        "    training_stats['val_loss'].append(val_loss)\r\n",
        "    training_stats['Validation Time'].append(validation_time)\r\n",
        "    training_stats['train_acc'].append(training_accuracy)\r\n",
        "    training_stats['val_acc'].append(val_accuracy)\r\n",
        "\r\n",
        "print(f'Best val accuracy: {best_accuracy}')\r\n",
        "  \r\n",
        "model.load_state_dict(torch.load('best_model_state.bin'))\r\n",
        "\r\n",
        "print(\"\")\r\n",
        "print(\"Training complete!\")\r\n",
        "\r\n",
        "print(\"Total training took {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "======== Epoch 1 / 4 ========\n",
            "Training on 536 batches...\n",
            "  Batch      50  of      536.    Elapsed: 0:00:02.  Remaining: 0:00:19\n",
            "  Batch     100  of      536.    Elapsed: 0:00:04.  Remaining: 0:00:17\n",
            "  Batch     150  of      536.    Elapsed: 0:00:06.  Remaining: 0:00:16\n",
            "  Batch     200  of      536.    Elapsed: 0:00:08.  Remaining: 0:00:14\n",
            "  Batch     250  of      536.    Elapsed: 0:00:11.  Remaining: 0:00:12\n",
            "  Batch     300  of      536.    Elapsed: 0:00:13.  Remaining: 0:00:10\n",
            "  Batch     350  of      536.    Elapsed: 0:00:15.  Remaining: 0:00:08\n",
            "  Batch     400  of      536.    Elapsed: 0:00:17.  Remaining: 0:00:06\n",
            "  Batch     450  of      536.    Elapsed: 0:00:19.  Remaining: 0:00:04\n",
            "  Batch     500  of      536.    Elapsed: 0:00:21.  Remaining: 0:00:02\n",
            "\n",
            "  Average training loss: 0.34\n",
            "  Training Accuracy: 0.71\n",
            "  Training epoch took: 0:00:23\n",
            "Creating Smart Batches from 1,070 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,070 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,070 samples\n",
            "\n",
            "     1,070 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Doing validation on 67 sentences...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:04\n",
            "  Batch      14  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:03\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      35  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:00\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n",
            "\n",
            "  Average validation loss: 0.34\n",
            "  Validation Accuracy: 0.73\n",
            "  Validation epoch took: 0:00:03\n",
            "\n",
            "======== Epoch 2 / 4 ========\n",
            "Creating Smart Batches from 8,564 examples with batch size 16...\n",
            "\n",
            "Tokenizing 8,564 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,800 samples.\n",
            "  Tokenized 2,700 samples.\n",
            "  Tokenized 3,600 samples.\n",
            "  Tokenized 4,500 samples.\n",
            "  Tokenized 5,400 samples.\n",
            "  Tokenized 6,300 samples.\n",
            "  Tokenized 7,200 samples.\n",
            "  Tokenized 8,100 samples.\n",
            "DONE.\n",
            "     8,564 samples\n",
            "\n",
            "     8,564 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 536 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Training on 536 batches...\n",
            "  Batch      50  of      536.    Elapsed: 0:00:02.  Remaining: 0:00:22\n",
            "  Batch     100  of      536.    Elapsed: 0:00:05.  Remaining: 0:00:20\n",
            "  Batch     150  of      536.    Elapsed: 0:00:07.  Remaining: 0:00:17\n",
            "  Batch     200  of      536.    Elapsed: 0:00:09.  Remaining: 0:00:15\n",
            "  Batch     250  of      536.    Elapsed: 0:00:11.  Remaining: 0:00:13\n",
            "  Batch     300  of      536.    Elapsed: 0:00:13.  Remaining: 0:00:10\n",
            "  Batch     350  of      536.    Elapsed: 0:00:15.  Remaining: 0:00:08\n",
            "  Batch     400  of      536.    Elapsed: 0:00:17.  Remaining: 0:00:06\n",
            "  Batch     450  of      536.    Elapsed: 0:00:19.  Remaining: 0:00:04\n",
            "  Batch     500  of      536.    Elapsed: 0:00:22.  Remaining: 0:00:02\n",
            "\n",
            "  Average training loss: 0.34\n",
            "  Training Accuracy: 0.73\n",
            "  Training epoch took: 0:00:24\n",
            "Creating Smart Batches from 1,070 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,070 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,070 samples\n",
            "\n",
            "     1,070 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Doing validation on 67 sentences...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:02\n",
            "  Batch      14  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      35  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:00\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n",
            "\n",
            "  Average validation loss: 0.34\n",
            "  Validation Accuracy: 0.74\n",
            "  Validation epoch took: 0:00:03\n",
            "\n",
            "======== Epoch 3 / 4 ========\n",
            "Creating Smart Batches from 8,564 examples with batch size 16...\n",
            "\n",
            "Tokenizing 8,564 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,800 samples.\n",
            "  Tokenized 2,700 samples.\n",
            "  Tokenized 3,600 samples.\n",
            "  Tokenized 4,500 samples.\n",
            "  Tokenized 5,400 samples.\n",
            "  Tokenized 6,300 samples.\n",
            "  Tokenized 7,200 samples.\n",
            "  Tokenized 8,100 samples.\n",
            "DONE.\n",
            "     8,564 samples\n",
            "\n",
            "     8,564 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 536 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Training on 536 batches...\n",
            "  Batch      50  of      536.    Elapsed: 0:00:02.  Remaining: 0:00:18\n",
            "  Batch     100  of      536.    Elapsed: 0:00:04.  Remaining: 0:00:19\n",
            "  Batch     150  of      536.    Elapsed: 0:00:07.  Remaining: 0:00:17\n",
            "  Batch     200  of      536.    Elapsed: 0:00:09.  Remaining: 0:00:15\n",
            "  Batch     250  of      536.    Elapsed: 0:00:11.  Remaining: 0:00:13\n",
            "  Batch     300  of      536.    Elapsed: 0:00:14.  Remaining: 0:00:11\n",
            "  Batch     350  of      536.    Elapsed: 0:00:15.  Remaining: 0:00:08\n",
            "  Batch     400  of      536.    Elapsed: 0:00:18.  Remaining: 0:00:06\n",
            "  Batch     450  of      536.    Elapsed: 0:00:20.  Remaining: 0:00:04\n",
            "  Batch     500  of      536.    Elapsed: 0:00:23.  Remaining: 0:00:02\n",
            "\n",
            "  Average training loss: 0.34\n",
            "  Training Accuracy: 0.73\n",
            "  Training epoch took: 0:00:25\n",
            "Creating Smart Batches from 1,070 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,070 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,070 samples\n",
            "\n",
            "     1,070 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Doing validation on 67 sentences...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:03\n",
            "  Batch      14  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:03\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      35  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:00\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n",
            "\n",
            "  Average validation loss: 0.34\n",
            "  Validation Accuracy: 0.74\n",
            "  Validation epoch took: 0:00:03\n",
            "\n",
            "======== Epoch 4 / 4 ========\n",
            "Creating Smart Batches from 8,564 examples with batch size 16...\n",
            "\n",
            "Tokenizing 8,564 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,800 samples.\n",
            "  Tokenized 2,700 samples.\n",
            "  Tokenized 3,600 samples.\n",
            "  Tokenized 4,500 samples.\n",
            "  Tokenized 5,400 samples.\n",
            "  Tokenized 6,300 samples.\n",
            "  Tokenized 7,200 samples.\n",
            "  Tokenized 8,100 samples.\n",
            "DONE.\n",
            "     8,564 samples\n",
            "\n",
            "     8,564 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 536 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Training on 536 batches...\n",
            "  Batch      50  of      536.    Elapsed: 0:00:02.  Remaining: 0:00:24\n",
            "  Batch     100  of      536.    Elapsed: 0:00:05.  Remaining: 0:00:22\n",
            "  Batch     150  of      536.    Elapsed: 0:00:07.  Remaining: 0:00:19\n",
            "  Batch     200  of      536.    Elapsed: 0:00:09.  Remaining: 0:00:15\n",
            "  Batch     250  of      536.    Elapsed: 0:00:12.  Remaining: 0:00:14\n",
            "  Batch     300  of      536.    Elapsed: 0:00:14.  Remaining: 0:00:11\n",
            "  Batch     350  of      536.    Elapsed: 0:00:16.  Remaining: 0:00:09\n",
            "  Batch     400  of      536.    Elapsed: 0:00:19.  Remaining: 0:00:06\n",
            "  Batch     450  of      536.    Elapsed: 0:00:21.  Remaining: 0:00:04\n",
            "  Batch     500  of      536.    Elapsed: 0:00:24.  Remaining: 0:00:02\n",
            "\n",
            "  Average training loss: 0.34\n",
            "  Training Accuracy: 0.74\n",
            "  Training epoch took: 0:00:25\n",
            "Creating Smart Batches from 1,070 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,070 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,070 samples\n",
            "\n",
            "     1,070 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Doing validation on 67 sentences...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:01\n",
            "  Batch      14  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:02\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      35  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:00\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n",
            "\n",
            "  Average validation loss: 0.34\n",
            "  Validation Accuracy: 0.75\n",
            "  Validation epoch took: 0:00:04\n",
            "Best val accuracy: 0.7457943925233644\n",
            "\n",
            "Training complete!\n",
            "Total training took 0:02:07 (h:mm:ss)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 450
        },
        "id": "GOmua_S1eXNM",
        "outputId": "44ab2a20-242b-4bf0-de97-88292f87dbb0"
      },
      "source": [
        "plot_training_history(training_stats)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3c-6bOIxVyrM"
      },
      "source": [
        "## Evaluating Performance Over Training Set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2QmKPNAeWHkZ",
        "outputId": "66bf23d3-a1a5-4c97-af18-6b17ca896b3b"
      },
      "source": [
        "(py_inputs, py_attn_masks, py_labels) = make_smart_batches(train_sentences, train_labels, batch_size ,tokenizer,max_input_length)\r\n",
        "\r\n",
        "y_pred , y_true = get_predictions(py_inputs, py_attn_masks, py_labels)\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Creating Smart Batches from 8,564 examples with batch size 16...\n",
            "\n",
            "Tokenizing 8,564 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,800 samples.\n",
            "  Tokenized 2,700 samples.\n",
            "  Tokenized 3,600 samples.\n",
            "  Tokenized 4,500 samples.\n",
            "  Tokenized 5,400 samples.\n",
            "  Tokenized 6,300 samples.\n",
            "  Tokenized 7,200 samples.\n",
            "  Tokenized 8,100 samples.\n",
            "DONE.\n",
            "     8,564 samples\n",
            "\n",
            "     8,564 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 536 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Predicting labels for 536 test batches...\n",
            "  Batch      50  of      536.    Elapsed: 0:00:02.  Remaining: 0:00:23\n",
            "  Batch     100  of      536.    Elapsed: 0:00:04.  Remaining: 0:00:19\n",
            "  Batch     150  of      536.    Elapsed: 0:00:07.  Remaining: 0:00:17\n",
            "  Batch     200  of      536.    Elapsed: 0:00:09.  Remaining: 0:00:15\n",
            "  Batch     250  of      536.    Elapsed: 0:00:11.  Remaining: 0:00:13\n",
            "  Batch     300  of      536.    Elapsed: 0:00:14.  Remaining: 0:00:11\n",
            "  Batch     350  of      536.    Elapsed: 0:00:16.  Remaining: 0:00:08\n",
            "  Batch     400  of      536.    Elapsed: 0:00:18.  Remaining: 0:00:06\n",
            "  Batch     450  of      536.    Elapsed: 0:00:20.  Remaining: 0:00:04\n",
            "  Batch     500  of      536.    Elapsed: 0:00:22.  Remaining: 0:00:02\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "uXsVn0wngZhv",
        "outputId": "c03a7cbc-5969-49b2-9dae-a15c2f78391e"
      },
      "source": [
        "print(classification_report(y_true, y_pred, target_names=class_names))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "                 precision    recall  f1-score   support\n",
            "\n",
            "    Hope_speech       0.60      0.00      0.01      1668\n",
            "Non_hope_speech       0.74      1.00      0.85      6205\n",
            "  not-malayalam       0.93      0.20      0.33       691\n",
            "\n",
            "       accuracy                           0.74      8564\n",
            "      macro avg       0.76      0.40      0.39      8564\n",
            "   weighted avg       0.73      0.74      0.64      8564\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 479
        },
        "id": "guClUdRhYPql",
        "outputId": "01d3bca0-af37-448a-eae1-30f2807dc7a9"
      },
      "source": [
        "cm = confusion_matrix(y_true, y_pred)\r\n",
        "show_confusion_matrix(cm, class_names)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VM2N-X1tV3il"
      },
      "source": [
        "## Evaluating Performance Over Validation Set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BnadNvtxWIYw",
        "outputId": "44459630-0d3c-4b1b-bb10-645d8114b023"
      },
      "source": [
        "(py_inputs, py_attn_masks, py_labels) = make_smart_batches(val_sentences, val_labels, batch_size ,tokenizer,max_input_length)\r\n",
        "\r\n",
        "y_pred , y_true = get_predictions(py_inputs, py_attn_masks, py_labels)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Creating Smart Batches from 1,070 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,070 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,070 samples\n",
            "\n",
            "     1,070 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Predicting labels for 67 test batches...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:02\n",
            "  Batch      14  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:03\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      35  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:00\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wAjlATBPgh7c",
        "outputId": "444cfb77-0c5c-4684-8bea-ed1fc287fed9"
      },
      "source": [
        "print(classification_report(y_true, y_pred, target_names=class_names))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "                 precision    recall  f1-score   support\n",
            "\n",
            "    Hope_speech       1.00      0.01      0.01       190\n",
            "Non_hope_speech       0.74      1.00      0.85       784\n",
            "  not-malayalam       0.88      0.15      0.25        96\n",
            "\n",
            "       accuracy                           0.75      1070\n",
            "      macro avg       0.87      0.38      0.37      1070\n",
            "   weighted avg       0.80      0.75      0.65      1070\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 479
        },
        "id": "7nELY57NYshD",
        "outputId": "05d76636-8a2e-4f01-9114-d8785d9552d8"
      },
      "source": [
        "cm = confusion_matrix(y_true, y_pred)\r\n",
        "show_confusion_matrix(cm, class_names)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j_Ev8kTSNT6L"
      },
      "source": [
        "## Evaluating Performance Over Test Set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "knzcVqvONT6N",
        "outputId": "06db5514-69d8-4e43-f685-7c1cda2b951f"
      },
      "source": [
        "(py_inputs, py_attn_masks, py_labels) = make_smart_batches(test_sentences, test_labels, batch_size ,tokenizer,max_input_length)\r\n",
        "\r\n",
        "y_pred , y_true = get_predictions(py_inputs, py_attn_masks, py_labels)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Creating Smart Batches from 1,071 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,071 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,071 samples\n",
            "\n",
            "     1,071 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n",
            "Predicting labels for 67 test batches...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:01\n",
            "  Batch      14  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:02\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      35  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:01\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "l6EJfj0TNT6R",
        "outputId": "2e7d8f0d-756b-41d8-b675-a29b5b24010b"
      },
      "source": [
        "print(classification_report(y_true, y_pred, target_names=class_names))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "                 precision    recall  f1-score   support\n",
            "\n",
            "    Hope_speech       0.00      0.00      0.00       194\n",
            "Non_hope_speech       0.74      0.99      0.85       776\n",
            "  not-malayalam       0.89      0.17      0.28       101\n",
            "\n",
            "       accuracy                           0.74      1071\n",
            "      macro avg       0.54      0.39      0.38      1071\n",
            "   weighted avg       0.62      0.74      0.64      1071\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "id": "ZE3EG9RTNT6S",
        "outputId": "f398fb63-b6a9-41b2-914c-469cf1794ecb"
      },
      "source": [
        "cm = confusion_matrix(y_true, y_pred)\r\n",
        "show_confusion_matrix(cm, class_names)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2fOJyH_c3ON2"
      },
      "source": [
        "## Saving the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3PXqG-VpurVS"
      },
      "source": [
        "torch.save(model,\"./drive/MyDrive/Codalab/HopeSpeech Detection For Equality,DiversityAnd Inclusion/Models/BERT_Mutlti_Malayalam\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a9M_91z1yj0F"
      },
      "source": [
        "## Loading the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "42ObCUU1QcsJ"
      },
      "source": [
        "model = torch.load(\r\n",
        "    './drive/MyDrive/HopeSpeech Detection For Equality,DiversityAnd Inclusion/Models/BERTForSequenceClassification_Mutlti_Malayalam_v2',\r\n",
        "    map_location=device\r\n",
        "    )"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sAaHpZdIzsnT",
        "outputId": "dbea6eee-1e96-4254-d082-a6a842d8fe89"
      },
      "source": [
        "model.cuda()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "BertForSequenceClassification(\n",
              "  (bert): BertModel(\n",
              "    (embeddings): BertEmbeddings(\n",
              "      (word_embeddings): Embedding(105879, 768, padding_idx=0)\n",
              "      (position_embeddings): Embedding(512, 768)\n",
              "      (token_type_embeddings): Embedding(2, 768)\n",
              "      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "      (dropout): Dropout(p=0.1, inplace=False)\n",
              "    )\n",
              "    (encoder): BertEncoder(\n",
              "      (layer): ModuleList(\n",
              "        (0): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (1): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (2): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (3): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (4): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (5): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (6): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (7): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (8): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (9): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (10): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (11): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "      )\n",
              "    )\n",
              "    (pooler): BertPooler(\n",
              "      (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "      (activation): Tanh()\n",
              "    )\n",
              "  )\n",
              "  (dropout): Dropout(p=0.1, inplace=False)\n",
              "  (classifier): Linear(in_features=768, out_features=3, bias=True)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 52
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RCo7olPPnYNz"
      },
      "source": [
        "## Making Predictions on Test Set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YkVQsHdKnYN0"
      },
      "source": [
        "test_df_path = join(dataset_dir,'malayalam_hope_test.csv')\r\n",
        "test_df = pd.read_csv(test_df_path,sep = '\\t')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        },
        "id": "nungZ-9dnYN0",
        "outputId": "e39dceb8-ff04-436a-bc01-d9c0fe6889d6"
      },
      "source": [
        "test_df"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>അതെ അണപൊട്ടി ഒഴുകുകയാണ്</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>കെവിൻ avde pettupoyi🤣🤣🤣</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>@my voice ullup illa daaa tayoli ath kond alle...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>@Albin Baiju</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>I respect you..</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1066</th>\n",
              "      <td>ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1067</th>\n",
              "      <td>@surya ishaan</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1068</th>\n",
              "      <td>ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1069</th>\n",
              "      <td>ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1070</th>\n",
              "      <td>Poorru  angodu chellu</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1071 rows × 1 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text\n",
              "0                               അതെ അണപൊട്ടി ഒഴുകുകയാണ്\n",
              "1                               കെവിൻ avde pettupoyi🤣🤣🤣\n",
              "2     @my voice ullup illa daaa tayoli ath kond alle...\n",
              "3                                          @Albin Baiju\n",
              "4                                       I respect you..\n",
              "...                                                 ...\n",
              "1066  ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...\n",
              "1067                                      @surya ishaan\n",
              "1068                ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ\n",
              "1069   ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...\n",
              "1070                              Poorru  angodu chellu\n",
              "\n",
              "[1071 rows x 1 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 68
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7jLPpljYQcsQ"
      },
      "source": [
        "## Loading `test_sentences`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QEVoAhgIQcsR"
      },
      "source": [
        "test_sentences = test_df['text'].values \r\n",
        "test_ids = test_df.index.values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "slovGuCRThW1",
        "outputId": "896028f4-4527-479f-dd76-b52c74afdfd6"
      },
      "source": [
        "test_ids"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([   0,    1,    2, ..., 1068, 1069, 1070])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 70
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gL7aCG2HTjv6",
        "outputId": "987c5f4d-7d48-4f36-9720-a81fe3f850bd"
      },
      "source": [
        "test_sentences"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['അതെ അണപൊട്ടി ഒഴുകുകയാണ്', 'കെവിൻ avde pettupoyi🤣🤣🤣',\n",
              "       '@my voice ullup illa daaa tayoli ath kond alle ninte ammaye njan panniyatha monee',\n",
              "       ..., 'ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ',\n",
              "       'ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...',\n",
              "       'Poorru  angodu chellu'], dtype=object)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 71
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "b9dgO1m3QcsU",
        "outputId": "031d1fdf-c46f-4a64-8650-9e5846392433"
      },
      "source": [
        "(py_inputs, py_attn_masks,py_ids) = make_smart_batches_on_test(test_sentences,test_ids, 16,tokenizer,max_input_length = 400)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Creating Smart Batches from 1,071 examples with batch size 16...\n",
            "\n",
            "Tokenizing 1,071 samples...\n",
            "  Tokenized 0 samples.\n",
            "  Tokenized 100 samples.\n",
            "  Tokenized 200 samples.\n",
            "  Tokenized 300 samples.\n",
            "  Tokenized 400 samples.\n",
            "  Tokenized 500 samples.\n",
            "  Tokenized 600 samples.\n",
            "  Tokenized 700 samples.\n",
            "  Tokenized 800 samples.\n",
            "  Tokenized 900 samples.\n",
            "  Tokenized 1,000 samples.\n",
            "DONE.\n",
            "     1,071 samples\n",
            "\n",
            "     1,071 samples after sorting\n",
            "\n",
            "Creating batches of size 16...\n",
            "\n",
            "  DONE - Selected 67 batches.\n",
            "\n",
            "Padding out sequences within each batch...\n",
            "  DONE.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vjsjqQpRQcsV"
      },
      "source": [
        "## Evaluating accuracy over test set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rwkx-30AQcsV",
        "outputId": "d2d16d92-bf8e-4184-b0d0-ea04f1a6fb9a"
      },
      "source": [
        "y_ids,y_preds  = get_predictions_test(py_inputs, py_attn_masks,py_ids)\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Predicting labels for 67 test batches...\n",
            "  Batch       7  of       67.    Elapsed: 0:00:00.  Remaining: 0:00:02\n",
            "  Batch      14  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      21  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:02\n",
            "  Batch      28  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      35  of       67.    Elapsed: 0:00:01.  Remaining: 0:00:01\n",
            "  Batch      42  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      49  of       67.    Elapsed: 0:00:02.  Remaining: 0:00:01\n",
            "  Batch      56  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:01\n",
            "  Batch      63  of       67.    Elapsed: 0:00:03.  Remaining: 0:00:00\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pDM-cnDHQcsW",
        "outputId": "b364feef-d836-44f8-c03d-fe3d5f4be317"
      },
      "source": [
        "print(y_preds)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[1 1 1 ... 0 0 1]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kWWWjz7eUPc7",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8f0f2176-6317-4380-8ba6-dd2cf9798529"
      },
      "source": [
        "print(y_ids)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[ 846  861  871 ...  741  178 1004]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u2ybbsAoQcsY",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5183909c-9ef7-4322-eece-f07c7ce56155"
      },
      "source": [
        "le.inverse_transform(y_preds)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['Non_hope_speech', 'Non_hope_speech', 'Non_hope_speech', ...,\n",
              "       'Hope_speech', 'Hope_speech', 'Non_hope_speech'], dtype=object)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 76
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "id": "kSMb6sCLQcsZ",
        "outputId": "10fd72ef-4397-4239-effd-01c6285b8c8b"
      },
      "source": [
        "sns.countplot(y =le.inverse_transform(y_preds))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f71aa5aa470>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 77
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4KXOXwH9U2MT",
        "outputId": "74497f91-0750-4ff8-e7a3-683a6b9420df"
      },
      "source": [
        "len(y_ids)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1071"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 78
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9-fkBgOxU4bX",
        "outputId": "d4e4aa8e-42d5-45cc-b3e2-9fd4035c48e4"
      },
      "source": [
        "len(le.inverse_transform(y_preds))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1071"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 79
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sWPDqogJrjIZ"
      },
      "source": [
        "## Creating Submission DataFrame"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o4Hb7RD1yHol"
      },
      "source": [
        "df_new = pd.DataFrame({\r\n",
        "    'id':y_ids,\r\n",
        "    'label':le.inverse_transform(y_preds)\r\n",
        "})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "twQAkVzcn4Lg"
      },
      "source": [
        "test_df['id'] = test_df.index"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        },
        "id": "qov36eWmqawv",
        "outputId": "75622337-0274-4f08-d1a7-fdeaaad2275e"
      },
      "source": [
        "df_new"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>id</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>846</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>861</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>871</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>953</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>990</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1066</th>\n",
              "      <td>355</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1067</th>\n",
              "      <td>585</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1068</th>\n",
              "      <td>741</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1069</th>\n",
              "      <td>178</td>\n",
              "      <td>Hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1070</th>\n",
              "      <td>1004</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1071 rows × 2 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "        id            label\n",
              "0      846  Non_hope_speech\n",
              "1      861  Non_hope_speech\n",
              "2      871  Non_hope_speech\n",
              "3      953  Non_hope_speech\n",
              "4      990  Non_hope_speech\n",
              "...    ...              ...\n",
              "1066   355  Non_hope_speech\n",
              "1067   585  Non_hope_speech\n",
              "1068   741      Hope_speech\n",
              "1069   178      Hope_speech\n",
              "1070  1004  Non_hope_speech\n",
              "\n",
              "[1071 rows x 2 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 82
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        },
        "id": "z0CEApHmqdLr",
        "outputId": "1039ba5f-a702-4582-8bfa-8f0a83fc3625"
      },
      "source": [
        "test_df"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>id</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>അതെ അണപൊട്ടി ഒഴുകുകയാണ്</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>കെവിൻ avde pettupoyi🤣🤣🤣</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>@my voice ullup illa daaa tayoli ath kond alle...</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>@Albin Baiju</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>I respect you..</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1066</th>\n",
              "      <td>ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...</td>\n",
              "      <td>1066</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1067</th>\n",
              "      <td>@surya ishaan</td>\n",
              "      <td>1067</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1068</th>\n",
              "      <td>ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ</td>\n",
              "      <td>1068</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1069</th>\n",
              "      <td>ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...</td>\n",
              "      <td>1069</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1070</th>\n",
              "      <td>Poorru  angodu chellu</td>\n",
              "      <td>1070</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1071 rows × 2 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text    id\n",
              "0                               അതെ അണപൊട്ടി ഒഴുകുകയാണ്     0\n",
              "1                               കെവിൻ avde pettupoyi🤣🤣🤣     1\n",
              "2     @my voice ullup illa daaa tayoli ath kond alle...     2\n",
              "3                                          @Albin Baiju     3\n",
              "4                                       I respect you..     4\n",
              "...                                                 ...   ...\n",
              "1066  ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...  1066\n",
              "1067                                      @surya ishaan  1067\n",
              "1068                ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ  1068\n",
              "1069   ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...  1069\n",
              "1070                              Poorru  angodu chellu  1070\n",
              "\n",
              "[1071 rows x 2 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 83
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        },
        "id": "vxReWWmxn7Hj",
        "outputId": "bf948560-c34c-4a16-b554-733764ab87b8"
      },
      "source": [
        "df_f = pd.merge(test_df,df_new,on = 'id')\r\n",
        "df_f"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>text</th>\n",
              "      <th>id</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>അതെ അണപൊട്ടി ഒഴുകുകയാണ്</td>\n",
              "      <td>0</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>കെവിൻ avde pettupoyi🤣🤣🤣</td>\n",
              "      <td>1</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>@my voice ullup illa daaa tayoli ath kond alle...</td>\n",
              "      <td>2</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>@Albin Baiju</td>\n",
              "      <td>3</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>I respect you..</td>\n",
              "      <td>4</td>\n",
              "      <td>not-malayalam</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1066</th>\n",
              "      <td>ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...</td>\n",
              "      <td>1066</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1067</th>\n",
              "      <td>@surya ishaan</td>\n",
              "      <td>1067</td>\n",
              "      <td>not-malayalam</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1068</th>\n",
              "      <td>ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ</td>\n",
              "      <td>1068</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1069</th>\n",
              "      <td>ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...</td>\n",
              "      <td>1069</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1070</th>\n",
              "      <td>Poorru  angodu chellu</td>\n",
              "      <td>1070</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1071 rows × 3 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text    id            label\n",
              "0                               അതെ അണപൊട്ടി ഒഴുകുകയാണ്     0  Non_hope_speech\n",
              "1                               കെവിൻ avde pettupoyi🤣🤣🤣     1  Non_hope_speech\n",
              "2     @my voice ullup illa daaa tayoli ath kond alle...     2  Non_hope_speech\n",
              "3                                          @Albin Baiju     3  Non_hope_speech\n",
              "4                                       I respect you..     4    not-malayalam\n",
              "...                                                 ...   ...              ...\n",
              "1066  ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...  1066  Non_hope_speech\n",
              "1067                                      @surya ishaan  1067    not-malayalam\n",
              "1068                ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ  1068  Non_hope_speech\n",
              "1069   ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...  1069  Non_hope_speech\n",
              "1070                              Poorru  angodu chellu  1070  Non_hope_speech\n",
              "\n",
              "[1071 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 84
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        },
        "id": "1kuASymYn9aV",
        "outputId": "60fc4fb9-1783-4dee-f856-a3f38e688d80"
      },
      "source": [
        "df_f = df_f[['id','text','label']]\r\n",
        "df_f"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>id</th>\n",
              "      <th>text</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>അതെ അണപൊട്ടി ഒഴുകുകയാണ്</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>കെവിൻ avde pettupoyi🤣🤣🤣</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>@my voice ullup illa daaa tayoli ath kond alle...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>@Albin Baiju</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>I respect you..</td>\n",
              "      <td>not-malayalam</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1066</th>\n",
              "      <td>1066</td>\n",
              "      <td>ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1067</th>\n",
              "      <td>1067</td>\n",
              "      <td>@surya ishaan</td>\n",
              "      <td>not-malayalam</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1068</th>\n",
              "      <td>1068</td>\n",
              "      <td>ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1069</th>\n",
              "      <td>1069</td>\n",
              "      <td>ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1070</th>\n",
              "      <td>1070</td>\n",
              "      <td>Poorru  angodu chellu</td>\n",
              "      <td>Non_hope_speech</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>1071 rows × 3 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "        id                                               text            label\n",
              "0        0                            അതെ അണപൊട്ടി ഒഴുകുകയാണ്  Non_hope_speech\n",
              "1        1                            കെവിൻ avde pettupoyi🤣🤣🤣  Non_hope_speech\n",
              "2        2  @my voice ullup illa daaa tayoli ath kond alle...  Non_hope_speech\n",
              "3        3                                       @Albin Baiju  Non_hope_speech\n",
              "4        4                                    I respect you..    not-malayalam\n",
              "...    ...                                                ...              ...\n",
              "1066  1066  ചേട്ടാ അതിന് ഇവിടെ ആരാണ് ഇന്ത്യൻ മൊബൈൽ റിവ്യൂ ...  Non_hope_speech\n",
              "1067  1067                                      @surya ishaan    not-malayalam\n",
              "1068  1068                ഏതാണ് ആ മൂലയിൽ ഇരിക്കുന്ന അപ്പൂപ്പൻ  Non_hope_speech\n",
              "1069  1069   ഏതാ ആ ചൊമന്ന ടി ഷർട്ടിന് മേൽ കോട്ടിട്ട വവ്വാൽ...  Non_hope_speech\n",
              "1070  1070                              Poorru  angodu chellu  Non_hope_speech\n",
              "\n",
              "[1071 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 85
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p3FtBVTVroJi"
      },
      "source": [
        "## Saving Submission DataFrame"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SzoT0p9rqrO-"
      },
      "source": [
        "path = './drive/MyDrive/Codalab/HopeSpeech Detection For Equality,DiversityAnd Inclusion/SubmissionFiles'"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LZmgZFM4qi7F"
      },
      "source": [
        "df_f.to_csv(join(path,'bert_multi_malayalam.csv'),index=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jW9g_qgMv7mq"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}