03-train-voltage-as-timesteps.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "03-train-voltage-as-timesteps-fix-error.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "XmXZqA0nbAYh"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "import pickle\n",
        "import pandas as pd\n",
        "import os\n",
        "import re\n",
        "import torch\n",
        "import torch.nn.functional as F\n",
        "from torch import nn\n",
        "from torch.utils.data import Dataset, DataLoader\n",
        "from torch import optim\n",
        "from scipy import signal\n",
        "import math\n",
        "from functools import lru_cache\n",
        "from torch import Tensor\n",
        "from typing import Type, Any, Callable, Union, List, Optional\n",
        "from sklearn import metrics\n",
        "import seaborn as sns"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1-0LyiSA9d7c"
      },
      "source": [
        "# transformer/lstm\n",
        "MODEL = 'lstm'"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E58imWjLNmzf"
      },
      "source": [
        "# model related hyperparams\n",
        "LSTM1_DIM = 128\n",
        "LSTM1_LAYERS = 1\n",
        "LSTM2_DIM = 64\n",
        "LSTM2_LAYERS = 1\n",
        "LSTM_REG_FC_DROPOUT=0.1"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CAGS4C1lKkHL"
      },
      "source": [
        "# dimension or transformer model\n",
        "TRANSFORMER1_DIM_FEEDFORWARD=2048\n",
        "TRANSFORMER1_DROPOUT=0.5\n",
        "TRANSFORMER2_DIM_FEEDFORWARD=2048\n",
        "TRANSFORMER2_DROPOUT=0.5\n",
        "TRANSFORMER_REG_FC_DROPOUT=0.5"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nVzDR0GI3yAB"
      },
      "source": [
        "# hyperparams\n",
        "WINDOW = 20\n",
        "# BATCH_SIZE = 64 # for colab cpu\n",
        "BATCH_SIZE = 32 # for colab gpu"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "X0n12dGkKsmY"
      },
      "source": [
        "# Data Preprocessing"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jQm_KHv_KdBT"
      },
      "source": [
        "download data from google drive"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yfHOh7eY1aCZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "be51ca53-3b7b-477a-f469-e2873e64efcd"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/gdrive')"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /gdrive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xhN-st_F1cvD"
      },
      "source": [
        "!unzip -q -d ./ /gdrive/My\\ Drive/cells.zip"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bQxQJRwq1eO3"
      },
      "source": [
        "DATA_DIR = \"cells\"\n",
        "files = os.listdir(DATA_DIR)\n",
        "\n",
        "# sort files by batch and cell number\n",
        "def sort_key(f):\n",
        "  match = re.match(r\"^b(\\d+)c(\\d+)\", f)\n",
        "  batch, cell = int(match[1]), int(match[2])\n",
        "  return batch * 1000 + cell\n",
        "\n",
        "files.sort(key=sort_key)\n",
        "files = np.array(files)"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HegSAdzU1kNB"
      },
      "source": [
        "# split data in the same way as paper\n",
        "numBat1 = len([f for f in files if f.startswith(\"b1\")])\n",
        "numBat2 = len([f for f in files if f.startswith(\"b2\")])\n",
        "numBat3 = len([f for f in files if f.startswith(\"b3\")])\n",
        "numBat = numBat1 + numBat2 + numBat3\n",
        "\n",
        "# test data\n",
        "test_ind = np.hstack((np.arange(0,(numBat1+numBat2),2),83))\n",
        "\n",
        "# train data\n",
        "train_ind = np.arange(1,(numBat1+numBat2-1),2)\n",
        "\n",
        "# the authors acquired this data after model developement\n",
        "secondary_test_ind = np.arange(numBat-numBat3,numBat)"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tGzohxlD1fsq"
      },
      "source": [
        "def get_cell_data(cell):\n",
        "  return pickle.load(open(os.path.join(DATA_DIR, cell), \"rb\"))"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EeS0ILh5Qixn"
      },
      "source": [
        "clean batch 2 data\n",
        "\n",
        "* spikes around ~250 cycle all cells\n",
        "* channel 3 computer restarted during end\n",
        "* channel 7, 21 thermocouple disconnected\n",
        "* channel 15, 16 thermocouple switched\n",
        "* channel 10 dies quickly its probably defective\n",
        "\n",
        "Since We dont have channel_id or cell_id    available in pkl files we use info from supplementary pdf to find those cells\n",
        "\n",
        "```\n",
        "Channel, Cell Id, Cycle Life, Recharge Policy, Split of corrupted cells obtained from supplemenary pdf.\n",
        "\n",
        "3    el150800460623 2237 3.6C(80%)-3.6C    test\n",
        "7    el150800460611 461  4.8C(80%)-4.8C    train\n",
        "21   el150800460603 489  4.65C(19%)-4.85C  train\n",
        "15   el150800460634 511  3.6C(30%)-6C      train\n",
        "16   el150800460451 561  3.6C(9%)-5C       test\n",
        "10   el150800460605 148  2C(10%)-6C        test\n",
        "```\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6R0dG_YyQlZ_"
      },
      "source": [
        "# index of defective cells saved to avoid computing again\n",
        "# channel_id, index in files, filename\n",
        "defective_cells = [\n",
        "  (3, 2, 'b1c2.pkl'),\n",
        "  (10, 42, 'b2c1.pkl'),\n",
        "  (15, 47, 'b2c6.pkl'),\n",
        "  (16, 48, 'b2c10.pkl'),\n",
        "  (21, 57, 'b2c21.pkl'),\n",
        "  (7, 61, 'b2c25.pkl')\n",
        "]"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jbH_1exF4iWm"
      },
      "source": [
        "def find_defective_cells():\n",
        "  # channel_id is not readable from python so we\n",
        "  # use combination of cycle_life and charing policy\n",
        "  # to identify cells\n",
        "\n",
        "  # info about corrupted cells obtained from supplemenary pdf\n",
        "  # channel_id, cycle_life, charging_policy\n",
        "  cell_info = [\n",
        "      (3, 2237, '3.6C(80%)-3.6C'),\n",
        "      (7, 461, '4.8C(80%)-4.8C'),\n",
        "      (21, 489, '4.65C(19%)-4.85C'),\n",
        "      (15, 511, '3.6C(30%)-6C'),\n",
        "      (16, 561, '3.6C(9%)-5C'),\n",
        "      (10, 148, '2C(10%)-6C'),\n",
        "  ]\n",
        "\n",
        "  defective_cells_found = []\n",
        "\n",
        "  for i, f in enumerate(files):\n",
        "    cell_data = get_cell_data(f)\n",
        "    cycle_life = cell_data['cycle_life'][0][0]\n",
        "    policy = cell_data['policy']\n",
        "    for cell in cell_info:\n",
        "      if cell[1] == cycle_life and cell[2] == policy:\n",
        "        defective_cells_found.append((cell[0], i, f))\n",
        "\n",
        "  defective_cells_found\n",
        "  return defective_cells_found\n",
        "\n",
        "if defective_cells is None:\n",
        "  defective_cells = find_defective_cells()"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TH2w7c2nQvTE"
      },
      "source": [
        "# remove 7, 21, 15, 16, 10 from train_ind, test_ind\n",
        "remove_channels = [7, 21, 15, 16, 10]\n",
        "remove_ind = [c[1] for c in defective_cells if c[0] in remove_channels]\n",
        "\n",
        "train_ind = np.array([ind for ind in train_ind if ind not in remove_ind])\n",
        "test_ind = np.array([ind for ind in test_ind if ind not in remove_ind])\n",
        "secondary_test_ind = np.array([ind for ind in secondary_test_ind if ind not in remove_ind])"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VbZ9xmrvQzTj"
      },
      "source": [
        "# change get_cell_data to remove spikes\n",
        "def get_cell_data(cell, apply_filter=True):\n",
        "  cell_data = pickle.load(open(os.path.join(DATA_DIR, cell), \"rb\"))\n",
        "  if apply_filter:\n",
        "    cell_data['summary']['QCharge'] = signal.medfilt(cell_data['summary']['QCharge'])\n",
        "    cell_data['summary']['QDischarge'] = signal.medfilt(cell_data['summary']['QDischarge'])\n",
        "  return cell_data"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AMcjabjJKzOm"
      },
      "source": [
        "Divide data into windows\n",
        "\n",
        "Our goal is to capture data for 20 consecutive charging cycles of cell, and predict remaining cycle life of the cell."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DrPbeqoFV4MC"
      },
      "source": [
        "!mkdir -p window/train\n",
        "!mkdir -p window/val\n",
        "!mkdir -p window/test"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kY2ZL0dVK1jn"
      },
      "source": [
        "def divide_into_window(split, files, window_size=20, delete_original=False, original_dir='cells'):\n",
        "  for i, f in enumerate(files):\n",
        "\n",
        "    cell_data = get_cell_data(f, False)\n",
        "    cycle_life = len(cell_data['summary']['Tavg'])\n",
        "    \n",
        "    start_idxs = list(range(1, cycle_life - window_size))\n",
        "    end_idxs = [start_idx + window_size for start_idx in start_idxs]\n",
        "\n",
        "    cell_file_name = os.path.basename(f).split(\".\")[0]\n",
        "\n",
        "    for start_idx, end_idx in zip(start_idxs, end_idxs):\n",
        "      \n",
        "      remaining_cycle_life = cycle_life - end_idx\n",
        "      assert remaining_cycle_life > 0, \"Negative Cycle Life\"\n",
        "\n",
        "      file_name = f\"{cell_file_name}_{start_idx}_{end_idx}_{remaining_cycle_life}.pkl\"\n",
        "\n",
        "      # Summary Features (1 per charging cycle)\n",
        "      tavg = cell_data['summary']['Tavg'][start_idx:end_idx]\n",
        "      ir = cell_data['summary']['IR'][start_idx:end_idx]\n",
        "      chargetime = cell_data['summary']['chargetime'][start_idx:end_idx]\n",
        "\n",
        "      assert len(tavg) == window_size, f\"invalid tavg length {len(tavg)}\" \n",
        "      assert len(ir) == window_size, f\"invalid ir length {len(ir)}\"\n",
        "      assert len(chargetime) == window_size, f\"invalid chargetime length {len(chargetime)}\"\n",
        "      \n",
        "      # Detail Features (1000 per charging cycle)\n",
        "      q = np.zeros((window_size, 1000))\n",
        "      t = np.zeros((window_size, 1000))\n",
        "      for i2, j in enumerate(range(start_idx, end_idx)):\n",
        "        qdlin = cell_data['cycles'][j]['Qdlin']\n",
        "        tdlin = cell_data['cycles'][j]['Tdlin']\n",
        "        q[i2] = qdlin\n",
        "        t[i2] = tdlin\n",
        "      \n",
        "      window_data = {\n",
        "          'ir': ir,\n",
        "          'tavg': tavg,\n",
        "          'chargetime': chargetime,\n",
        "          'qdlin': q,\n",
        "          'tdlin': t,\n",
        "          'cycle_life': cycle_life,\n",
        "          'cycle_life_remaining': remaining_cycle_life\n",
        "      }\n",
        "\n",
        "      with open(os.path.join('window', split, file_name), \"wb\") as window_file:\n",
        "        pickle.dump(window_data, window_file)\n",
        "\n",
        "    # delete original file to save space \n",
        "    if delete_original:\n",
        "      os.remove(os.path.join(original_dir, f))"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P57KyATuMe1z"
      },
      "source": [
        "divide_into_window('train',\n",
        "                   files[train_ind], \n",
        "                   window_size=WINDOW,\n",
        "                   original_dir=DATA_DIR, \n",
        "                   delete_original=True)"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aZUMfem8d6zM"
      },
      "source": [
        "divide_into_window('val', \n",
        "                   files[test_ind],\n",
        "                   window_size=WINDOW,\n",
        "                   original_dir=DATA_DIR, \n",
        "                   delete_original=True)"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DOyEOSc4d-ne"
      },
      "source": [
        "# divide_into_window('test', \n",
        "#                    files[secondary_test_ind], \n",
        "#                    original_dir=DATA_DIR, \n",
        "#                    delete_original=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w8ZP_Ch7Yej7"
      },
      "source": [
        "def get_files(split):\n",
        "  file_names = os.listdir(os.path.join('window', split))\n",
        "  return [os.path.join('window', split, f) for f in file_names]"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "g4_2SfP5Y977"
      },
      "source": [
        "train_files = get_files('train')\n",
        "val_files = get_files('val')\n",
        "# test_files = get_files('test')"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Yo4mbw-OZCaY",
        "outputId": "b560cdbe-3bb5-4cf7-9c94-b2f7fa073a22"
      },
      "source": [
        "print(\"train samples\", len(train_files))\n",
        "print(\"validation samples\", len(val_files))\n",
        "# print(len(\"test samples\", test_files))"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "train samples 25830\n",
            "validation samples 29454\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mqBWqLHXQ0PK"
      },
      "source": [
        "# Dataset and Dataloader"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FcX9SXWkhpUv"
      },
      "source": [
        "class CycleLifeDataset(Dataset):\n",
        "  def __init__(self, files, window=20):\n",
        "    self.files = files\n",
        "\n",
        "  def __len__(self):\n",
        "    return len(self.files)\n",
        "\n",
        "  def __getitem__(self, idx):\n",
        "    # min, max, mean, std\n",
        "    tdlin_stats = [30, 39, 32.16, 0.80]\n",
        "    qdlin_stats = [0.002, 1.043, 0.68, 0.40]\n",
        "    chargetime_stats = [10, 14, 10.95, 1.14]\n",
        "\n",
        "    with open(self.files[idx], \"rb\") as f: \n",
        "      window_data = pickle.load(f)\n",
        "      \n",
        "      # qdlin, tdlin shape (window, 1000)\n",
        "      # ir, tavg, chargetime shape (window, )\n",
        "      qdlin = self._normalize(window_data['qdlin'], qdlin_stats) \n",
        "      tdlin = self._normalize(window_data['tdlin'], tdlin_stats)\n",
        "      tavg = window_data['tavg'] / 39\n",
        "      ir = window_data['ir'] / 0.02\n",
        "      chargetime = self._normalize(window_data['chargetime'], chargetime_stats)\n",
        "\n",
        "      # Detailed Data (1000 values per charging cycle)\n",
        "      x1 = np.stack([qdlin, tdlin]) # (2, window, 1000)\n",
        "      x1 = torch.tensor(x1, dtype=torch.float32).permute((1, 0, 2)) # (window, 2, 1000)\n",
        "\n",
        "      # Summary Data (1 value per charging cycle)\n",
        "      x2 = np.stack([tavg, ir, chargetime]) # (3, window)\n",
        "      x2 = torch.tensor(x2, dtype=torch.float32).permute((1, 0)) # (window, 3)\n",
        "\n",
        "      # remaining cycle life\n",
        "      cycle_life_remaining = window_data['cycle_life_remaining'] + 1e-6\n",
        "      y = torch.log(torch.tensor([cycle_life_remaining], dtype=torch.float32))\n",
        "\n",
        "      # ((window, 2, 1000), (window, 3)), (1, )\n",
        "      return (x1, x2), y\n",
        "\n",
        "  def _normalize(self, data, stats):\n",
        "    return (np.clip(data, stats[0], stats[1]) - stats[2]) / stats[3]"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qr__gdnXqy-1"
      },
      "source": [
        "train_ds = CycleLifeDataset(train_files, window=WINDOW)\n",
        "val_ds = CycleLifeDataset(val_files, window=WINDOW)\n",
        "# test_ds = CycleLifeDataset(files[secondary_test_ind], window=WINDOW)\n",
        "\n",
        "train_dl = DataLoader(train_ds, batch_size=BATCH_SIZE, shuffle=True)\n",
        "val_dl = DataLoader(val_ds, batch_size=BATCH_SIZE)\n",
        "# test_dl = DataLoader(test_ds, batch_size=BATCH_SIZE)"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zkitl_i8LBKx"
      },
      "source": [
        "device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")\n",
        "\n",
        "def to_device(data, device):\n",
        "  \"\"\"\n",
        "  loads data onto device\n",
        "  \"\"\"\n",
        "  if isinstance(data, (list, tuple)):\n",
        "    return [to_device(d, device) for d in data]\n",
        "  if isinstance(data, dict):\n",
        "    return {k: to_device(v, device) for (k, v) in data.items()}\n",
        "  return data.to(device, non_blocking=True)\n",
        "\n",
        "class DeviceDataLoader():\n",
        "  def __init__(self, dl, device):\n",
        "    \"\"\"\n",
        "    wraps a pytorch dataloader\n",
        "    for each batch that dataloader yields\n",
        "    it loads that batch onto device\n",
        "    \"\"\"\n",
        "    self.dl = dl\n",
        "    self.device = device\n",
        "  \n",
        "  def __iter__(self):\n",
        "    for data in self.dl:\n",
        "      yield to_device(data, self.device)\n",
        "\n",
        "  def __len__(self):\n",
        "    return len(self.dl)"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uA0ttWBGLHWq"
      },
      "source": [
        "train_dl = DeviceDataLoader(train_dl, device)\n",
        "val_dl = DeviceDataLoader(val_dl, device)"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "293JtXi55T9Z"
      },
      "source": [
        "# expected shapes\n",
        "x1_shape = torch.Size((WINDOW, 2, 1000))\n",
        "x2_shape = torch.Size((WINDOW, 3))\n",
        "y_shape = torch.Size((1, ))\n",
        "\n",
        "sample = train_ds[0]\n",
        "x1, x2 = sample[0]\n",
        "y = sample[1]\n",
        "\n",
        "assert x1.shape == x1_shape, \"Invalid x1 shape\"\n",
        "assert x2.shape == x2_shape, \"Invalid x2 shape\"\n",
        "assert y.shape == y_shape, \"Invalid y shape\""
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "obTBS5xD6XcU"
      },
      "source": [
        "xb, yb = next((iter(train_dl)))\n",
        "\n",
        "assert xb[0].shape == torch.Size((BATCH_SIZE, *x1_shape)), \"Invalid xb[0] shape\"\n",
        "assert xb[1].shape == torch.Size((BATCH_SIZE, *x2_shape)), \"Invalid xb[1] shape\"\n",
        "assert yb.shape == torch.Size((BATCH_SIZE, *y_shape)), \"Invalid yb shape\""
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "emmGVWpGYsoZ"
      },
      "source": [
        "# Models\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a0UDSgS8CGJX"
      },
      "source": [
        "Our task is to look at 20 consecutive charging cycles of cell to predict remaining cycle life of the cell. Our data has 2 types of features \n",
        "\n",
        "**1. Detailed Features**\n",
        "We have 1000 values of these features per charging cycle. We have measurements for 1000 voltage steps from 2v to 3.6v. We have 2 of these features. So shape of our is `(batch, 20 charging cycles, 1000 measurements, 2 features)`\n",
        "\n",
        "**2. Summary Features** We have one measurement of these feature per charging cycle. So shape of our data is `(batch, 20 charging cycles, 3 features)`\n",
        "\n",
        "These models extract features from both detailed measurements and summary measurements and pass it to fully connected layer for regression.\n",
        "\n",
        "These models treat Detailed features as `(batch, 1000 voltage steps, 2 features * 20 values)`. Doing this allows us to have just 1000 timesteps instead of 20000 timesteps.\n",
        "\n",
        "Another approach would be to loop a model over 20 charging cycles so we have single vector per cycle. Combine that with summary features. And then use lstm over 20 charging cycles. Followed by final fully connected layer."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6V6VJMzrOeFa"
      },
      "source": [
        "# since most cells last around 1000-2000 charging cycles\n",
        "# we'll clip relu to max 8 (e^8 ~2900 cycles)\n",
        "\n",
        "class ReLUX(nn.Module):\n",
        "    def __init__(self, max_value: float=1.0):\n",
        "        super(ReLUX, self).__init__()\n",
        "        self.max_value = float(max_value)\n",
        "        self.scale     = 6.0/self.max_value\n",
        "\n",
        "    def forward(self, x):\n",
        "        return F.relu6(x * self.scale) / (self.scale)"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-6Cdqhmjwh2o"
      },
      "source": [
        "### LSTM Regressor\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CBOIQd8i9UlX"
      },
      "source": [
        "class LSTMRegressor(nn.Module):\n",
        "  def __init__(self, window=20, \n",
        "               lstm1_dim=128, lstm1_layers=1, \n",
        "               lstm2_dim=128, lstm2_layers=1,\n",
        "               fc_dropout=0):\n",
        "\n",
        "    super().__init__()\n",
        "\n",
        "    self.window = window\n",
        "\n",
        "    # lstm for detailed features\n",
        "    # which have 1000 values over a cycle\n",
        "    self.lstm1 = nn.LSTM(2*window, lstm1_dim, num_layers=lstm1_layers)\n",
        "\n",
        "    # lstm for summary features\n",
        "    # which have 1 values over a cycle\n",
        "    self.lstm2 = nn.LSTM(3, lstm2_dim, num_layers=lstm2_layers)\n",
        "    \n",
        "    # combine features from both lstm\n",
        "    self.fc = nn.Sequential(\n",
        "      nn.Dropout(fc_dropout),\n",
        "      nn.Linear(lstm1_dim + lstm2_dim, 1),\n",
        "      ReLUX(8), \n",
        "    )\n",
        "\n",
        "  def _reshape_inputs(self, x):\n",
        "    # x_detailed float32 (batch, window, 2, 1000)\n",
        "    # x_summary float32 (batch, window, 3)\n",
        "    x_detailed, x_summary = x\n",
        "\n",
        "    # we see input as window * 2 features over 1000 timesteps\n",
        "    # (1000, batch, window*4)\n",
        "    x_detailed = x_detailed.permute(3, 0, 1, 2).flatten(start_dim=2)\n",
        "\n",
        "    # (window, batch, 3)\n",
        "    x_summary = x_summary.permute(1, 0, 2)\n",
        "    \n",
        "    return x_detailed, x_summary\n",
        "\n",
        "  def forward(self, x):\n",
        "\n",
        "    # x_detailed (1000, batch, window*4)\n",
        "    # x_summary (window, batch, 3)\n",
        "    x_detailed, x_summary = self._reshape_inputs(x)\n",
        "\n",
        "    # (batch, lstm1_dim)\n",
        "    x_detailed_out = self.lstm1(x_detailed)[0][-1]\n",
        "\n",
        "    # (batch, lstm2_dim)\n",
        "    x_summary_out = self.lstm2(x_summary)[0][-1]\n",
        "\n",
        "    # (batch, lstm1_dim + lstm2_dim)\n",
        "    x_combined = torch.cat((x_summary_out, x_detailed_out), dim=1)\n",
        "\n",
        "    # (batch, 1)\n",
        "    out = self.fc(x_combined)\n",
        "\n",
        "    return out"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3jZFePWUy6DN"
      },
      "source": [
        "### Transformer regressor"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0woDSeDxIINq"
      },
      "source": [
        "class PositionalEncoding(nn.Module):\n",
        "\n",
        "    def __init__(self, d_model: int, dropout: float = 0.1, max_len: int = 5000):\n",
        "        super().__init__()\n",
        "        self.dropout = nn.Dropout(p=dropout)\n",
        "\n",
        "        position = torch.arange(max_len).unsqueeze(1)\n",
        "        div_term = torch.exp(torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))\n",
        "        pe = torch.zeros(max_len, 1, d_model)\n",
        "        pe[:, 0, 0::2] = torch.sin(position * div_term)\n",
        "        pe[:, 0, 1::2] = torch.cos(position * div_term)\n",
        "        self.register_buffer('pe', pe)\n",
        "\n",
        "    def forward(self, x):\n",
        "        \"\"\"\n",
        "        Args:\n",
        "            x: Tensor, shape [seq_len, batch_size, embedding_dim]\n",
        "        \"\"\"\n",
        "        x = x + self.pe[:x.size(0)]\n",
        "        return self.dropout(x)\n",
        "\n",
        "class TransformerRegressor(nn.Module):\n",
        "  def __init__(self, window=20, \n",
        "               transformer1_dim_feedforward=2048,\n",
        "               transformer2_dim_feedforward=2048,\n",
        "               transformer1_fc_dropout=0.1,\n",
        "               transformer2_fc_dropout=0.1,\n",
        "               reg_fc_dropout=0.5):\n",
        "\n",
        "    super().__init__()\n",
        "\n",
        "    self.window = window\n",
        "\n",
        "    self.positional1 = PositionalEncoding(\n",
        "        self.window*2,\n",
        "        max_len=1000,\n",
        "    )\n",
        "\n",
        "    self.transformer1 = nn.TransformerEncoderLayer(\n",
        "        d_model=self.window*2,\n",
        "        nhead=2,\n",
        "        dim_feedforward=transformer1_dim_feedforward,\n",
        "        dropout=transformer1_fc_dropout\n",
        "    )\n",
        "\n",
        "    self.positional2 = PositionalEncoding(\n",
        "        4,\n",
        "        max_len=self.window,\n",
        "    )\n",
        "\n",
        "    self.transformer2 = nn.TransformerEncoderLayer(\n",
        "        d_model=4,\n",
        "        nhead=1,\n",
        "        dim_feedforward=transformer2_dim_feedforward,\n",
        "        dropout=transformer2_fc_dropout\n",
        "    )\n",
        "  \n",
        "    # combine features from both lstm\n",
        "    self.fc = nn.Sequential(\n",
        "      nn.Dropout(reg_fc_dropout),\n",
        "      nn.Linear(self.window * 2 + 4, 1),\n",
        "      ReLUX(8),\n",
        "    )\n",
        "\n",
        "    self.device = device\n",
        "\n",
        "  def _reshape_inputs(self, x):\n",
        "    # x_detailed float32 (batch, window, 2, 1000)\n",
        "    # x_summary float32 (batch, window, 3)\n",
        "    x_detailed, x_summary = x\n",
        "\n",
        "    # we see input as window * 2 features over 1000 timesteps\n",
        "    # (1000, batch, window*4)\n",
        "    x_detailed = x_detailed.permute(3, 0, 1, 2).flatten(start_dim=2)\n",
        "\n",
        "    # (window, batch, 4)\n",
        "    x_summary = F.pad(x_summary.permute(1, 0, 2), (1, 0))\n",
        "\n",
        "    return x_detailed, x_summary\n",
        "\n",
        "  def forward(self, x):\n",
        "    # x_detailed (1000, batch, window*4)\n",
        "    # x_summary (window, batch, 4)\n",
        "    x_detailed, x_summary = self._reshape_inputs(x)\n",
        "\n",
        "    x_detailed_out = self.transformer1(self.positional1(x_detailed)).mean(dim=0)\n",
        "    x_summary_out = self.transformer2(self.positional2(x_summary)).mean(dim=0)\n",
        "\n",
        "    x_combined = torch.cat((x_summary_out, x_detailed_out), dim=1)\n",
        "\n",
        "    out = self.fc(x_combined)\n",
        "\n",
        "    return out"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jAvDeEPrrzaI"
      },
      "source": [
        "def make_model():\n",
        "  model = None\n",
        "  if MODEL == 'transformer':\n",
        "    model = TransformerRegressor(window=WINDOW,\n",
        "                                 transformer1_dim_feedforward=TRANSFORMER1_DIM_FEEDFORWARD, \n",
        "                                 transformer2_dim_feedforward=TRANSFORMER2_DIM_FEEDFORWARD,\n",
        "                                 transformer1_fc_dropout=TRANSFORMER1_DROPOUT,\n",
        "                                 transformer2_fc_dropout=TRANSFORMER2_DROPOUT,\n",
        "                                 reg_fc_dropout=TRANSFORMER_REG_FC_DROPOUT)\n",
        "  if MODEL == 'lstm':\n",
        "    model = LSTMRegressor(window=WINDOW,\n",
        "                          lstm1_dim=LSTM1_DIM, \n",
        "                          lstm1_layers=LSTM1_LAYERS,\n",
        "                          lstm2_dim=LSTM2_DIM, \n",
        "                          lstm2_layers=LSTM2_LAYERS,\n",
        "                          fc_dropout=LSTM_REG_FC_DROPOUT)\n",
        "\n",
        "  return to_device(model, device)"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nfmoNL51KtQu"
      },
      "source": [
        "model = make_model()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1VzJFrXgCMU4"
      },
      "source": [
        "input = (torch.rand((BATCH_SIZE, *x1_shape)), torch.rand((BATCH_SIZE, *x2_shape)))\n",
        "\n",
        "assert model(to_device(input, device)).shape == torch.Size((BATCH_SIZE, 1)), \"Invalid Model Output Shape\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oT0iIcJOUQ35"
      },
      "source": [
        "# Loss function and metrics"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pREioOqk6LBt"
      },
      "source": [
        "# metrics\n",
        "loss_fn = nn.MSELoss()\n",
        "\n",
        "def mse(ypred, ytrue):\n",
        "  \"\"\"\n",
        "  model predicts ln(y), so we compute e^y\n",
        "  to bring y to original scale before calculating rmse\n",
        "  \"\"\"\n",
        "  return F.mse_loss(torch.exp(ypred), torch.exp(ytrue))\n",
        "\n",
        "def mae(ypred, ytrue):\n",
        "  \"\"\"\n",
        "  model predicts ln(y), so we compute e^y\n",
        "  to bring y to original scale before calculating mae\n",
        "  \"\"\"\n",
        "  ypred = torch.exp(ypred)\n",
        "  ytrue = torch.exp(ytrue)\n",
        "  return torch.mean(torch.abs(ypred-ytrue))"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OW3wLJtQUWCD"
      },
      "source": [
        "# Training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Sgy60O61UYzx"
      },
      "source": [
        "def epoch_loop(model, \n",
        "               dataloader, \n",
        "               loss_fn, \n",
        "               metric_fns, \n",
        "               optimizer=None, \n",
        "               train=False,):\n",
        "  \"\"\"\n",
        "  train or validation loop\n",
        "  \"\"\"\n",
        "\n",
        "  assert not train or optimizer is not None, \"optimizer cannot be null while training\"\n",
        "\n",
        "  if train:\n",
        "    model.train()\n",
        "  else:\n",
        "    model.eval()\n",
        "\n",
        "  loss_sum = 0  \n",
        "  metric_sum = np.zeros((len(metric_fns)))\n",
        "  n_samples = 0\n",
        "\n",
        "  print()\n",
        "  for i, batch in enumerate(dataloader):\n",
        "\n",
        "    xb, yb = batch\n",
        "\n",
        "    if train:\n",
        "      # reset gradients\n",
        "      optimizer.zero_grad()\n",
        "\n",
        "    # forward\n",
        "    ypred = model(xb)\n",
        "    loss = loss_fn(ypred, yb)\n",
        "\n",
        "    if train:\n",
        "      # backward \n",
        "      loss.backward()\n",
        "\n",
        "      # update weights\n",
        "      optimizer.step()\n",
        "\n",
        "    with torch.no_grad():\n",
        "      batch_size = len(xb)\n",
        "\n",
        "      # update number of samples\n",
        "      n_samples += batch_size\n",
        "\n",
        "      # accumulate loss\n",
        "      loss_sum  += loss.item() * batch_size\n",
        "\n",
        "      # accumulate metrics\n",
        "      for j, metric_fn in enumerate(metric_fns):\n",
        "        metric_sum[j] += metric_fn(ypred, yb) * batch_size\n",
        "      if (i % 10) == 0:\n",
        "        print(f\"Step {i} loss: {loss.item():0.4f}\")\n",
        "\n",
        "  print()\n",
        "\n",
        "  metric_avg = metric_sum / n_samples\n",
        "  loss_avg = loss_sum / n_samples\n",
        "\n",
        "  # return loss and metrics\n",
        "  return (loss_avg, *metric_avg)"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uaOMlgqOYcSl"
      },
      "source": [
        "class Trainer():\n",
        "  def __init__(self, model, optimizer, \n",
        "               train_dl, val_dl, \n",
        "               scheduler=None, \n",
        "               best_val_score=1e4,\n",
        "               best_weights=None):\n",
        "\n",
        "    self.model = model\n",
        "\n",
        "    self.optimizer = optimizer\n",
        "\n",
        "    self.scheduler = scheduler\n",
        "\n",
        "    self.best_val_score = best_val_score\n",
        "    self.best_weights = best_weights\n",
        "\n",
        "    self.history = {\n",
        "        'train' : {\n",
        "            'loss': [],\n",
        "            'rmse': [],\n",
        "            'mae': [],\n",
        "        },\n",
        "        'val' : {\n",
        "            'loss': [],\n",
        "            'rmse': [],\n",
        "            'mae': [],\n",
        "        },\n",
        "    }\n",
        "\n",
        "    self.train_dl = train_dl\n",
        "\n",
        "    self.val_dl = val_dl\n",
        "\n",
        "    # epochs completed\n",
        "    self.epoch = 0\n",
        "\n",
        "    self.loss_fn = nn.MSELoss()\n",
        "\n",
        "    self.metric_fns = [mse, mae]\n",
        "  \n",
        "  def train(self, n_epochs):\n",
        "\n",
        "    start_epoch = self.epoch + 1  \n",
        "    end_epoch = self.epoch + n_epochs\n",
        "\n",
        "    for epoch in range(start_epoch, end_epoch+1):\n",
        "      print(f\"Epoch {epoch}/{end_epoch}\")\n",
        "      # train epoch\n",
        "      train_res = epoch_loop(self.model, \n",
        "                             self.train_dl, \n",
        "                             self.loss_fn, \n",
        "                             self.metric_fns, \n",
        "                             train=True, \n",
        "                             optimizer=self.optimizer)\n",
        "\n",
        "      self.log('train', train_res)\n",
        "\n",
        "      # validate epoch\n",
        "      val_res = epoch_loop(\n",
        "          self.model, \n",
        "          self.train_dl,\n",
        "          self.loss_fn, \n",
        "          self.metric_fns,\n",
        "          train=False)\n",
        "\n",
        "      # print again so its easier to see\n",
        "      print()\n",
        "      self.log('train', train_res, save_history=False)\n",
        "      self.log('val', val_res)\n",
        "\n",
        "      self.save_best(val_res)\n",
        "  \n",
        "      if self.scheduler is not None:\n",
        "        self.scheduler.step(val_res[0])\n",
        "\n",
        "      print(\"\\n\")\n",
        "\n",
        "      self.epoch = epoch\n",
        "  \n",
        "  def save_best(self, val_res):\n",
        "    loss_val, mse_val, mae_val = val_res\n",
        "    rmse_val = math.sqrt(mse_val)\n",
        "\n",
        "    current_score = rmse_val\n",
        "    best_score = self.best_val_score\n",
        "    if current_score < best_score:\n",
        "      filename = f'best_model_loss_{loss_val:0.4f}_rmse_{rmse_val:0.4f}_mae_{mae_val:0.4f}.pth'\n",
        "      torch.save(self.model.state_dict(), filename)\n",
        "      self.best_weights = filename\n",
        "      self.best_val_score = current_score\n",
        "\n",
        "  def restore_best(self):\n",
        "    if self.model is not None and self.best_weights is not None:\n",
        "      self.model.load_state_dict(torch.load(self.best_weights))\n",
        "   \n",
        "  def log(self, split, result, save_history=True):\n",
        "    loss_val, mse_val, mae_val = result\n",
        "    rmse_val = math.sqrt(mse_val)\n",
        "\n",
        "    if save_history:\n",
        "      self.history[split]['loss'] = loss_val \n",
        "      self.history[split]['rmse'] = rmse_val\n",
        "      self.history[split]['mae'] = mae_val\n",
        "\n",
        "    print(f\"{split}_loss: {loss_val:0.4f} {split}_rmse: {rmse_val:0.4f} {split}_mae: {mae_val:0.4f}\")"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2C7a11s29t07"
      },
      "source": [
        "seed = 42\n",
        "torch.manual_seed(seed)\n",
        "np.random.seed(seed)\n",
        "torch.cuda.manual_seed_all(seed)"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bCzu70SHK_Cr"
      },
      "source": [
        "model = make_model()"
      ],
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Gm9u1_ww1Cf5",
        "outputId": "5ffcbc68-74ee-4847-834e-f8a03324f321"
      },
      "source": [
        "n_params = sum(param.numel() for param in model.parameters())\n",
        "print(f\"{MODEL} has {n_params} params\")"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "lstm has 104897 params\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wOXtNxfbfxba"
      },
      "source": [
        "params = model.parameters()\n",
        "\n",
        "optimizer = optim.AdamW(params, lr=7e-5)\n",
        "\n",
        "scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(\n",
        "    optimizer, \n",
        "    factor=0.5, \n",
        "    patience=1,\n",
        "    min_lr=1e-6,\n",
        ")"
      ],
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_qn81cV5FbhC"
      },
      "source": [
        "trainer = Trainer(model, optimizer, train_dl, val_dl, scheduler=scheduler) "
      ],
      "execution_count": 24,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zzJL2GeJTfzK",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c4cc51fb-11f0-4177-8ebf-72c57aa5a17f"
      },
      "source": [
        "trainer.train(10)"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "\n",
            "Step 0 loss: 29.4688\n",
            "Step 10 loss: 29.7742\n",
            "Step 20 loss: 25.6004\n",
            "Step 30 loss: 26.5142\n",
            "Step 40 loss: 22.1811\n",
            "Step 50 loss: 23.9362\n",
            "Step 60 loss: 19.7256\n",
            "Step 70 loss: 20.4846\n",
            "Step 80 loss: 12.0979\n",
            "Step 90 loss: 13.0979\n",
            "Step 100 loss: 12.5153\n",
            "Step 110 loss: 10.4855\n",
            "Step 120 loss: 9.0415\n",
            "Step 130 loss: 7.0476\n",
            "Step 140 loss: 5.6511\n",
            "Step 150 loss: 7.2164\n",
            "Step 160 loss: 7.1205\n",
            "Step 170 loss: 4.7856\n",
            "Step 180 loss: 6.4632\n",
            "Step 190 loss: 3.4014\n",
            "Step 200 loss: 4.0569\n",
            "Step 210 loss: 2.4487\n",
            "Step 220 loss: 2.6986\n",
            "Step 230 loss: 2.5775\n",
            "Step 240 loss: 1.1153\n",
            "Step 250 loss: 1.9580\n",
            "Step 260 loss: 0.5007\n",
            "Step 270 loss: 2.1893\n",
            "Step 280 loss: 1.3569\n",
            "Step 290 loss: 1.3090\n",
            "Step 300 loss: 1.1780\n",
            "Step 310 loss: 0.7198\n",
            "Step 320 loss: 0.9251\n",
            "Step 330 loss: 1.6360\n",
            "Step 340 loss: 0.5085\n",
            "Step 350 loss: 1.0277\n",
            "Step 360 loss: 0.5730\n",
            "Step 370 loss: 0.9503\n",
            "Step 380 loss: 0.6226\n",
            "Step 390 loss: 0.8184\n",
            "Step 400 loss: 0.6012\n",
            "Step 410 loss: 0.4412\n",
            "Step 420 loss: 1.1287\n",
            "Step 430 loss: 0.6694\n",
            "Step 440 loss: 0.7187\n",
            "Step 450 loss: 0.9216\n",
            "Step 460 loss: 0.8771\n",
            "Step 470 loss: 0.4472\n",
            "Step 480 loss: 0.2970\n",
            "Step 490 loss: 0.7837\n",
            "Step 500 loss: 0.4054\n",
            "Step 510 loss: 0.5711\n",
            "Step 520 loss: 0.4358\n",
            "Step 530 loss: 0.3613\n",
            "Step 540 loss: 0.7199\n",
            "Step 550 loss: 0.3901\n",
            "Step 560 loss: 0.3685\n",
            "Step 570 loss: 0.2188\n",
            "Step 580 loss: 0.3943\n",
            "Step 590 loss: 0.3837\n",
            "Step 600 loss: 0.4921\n",
            "Step 610 loss: 0.4497\n",
            "Step 620 loss: 0.8205\n",
            "Step 630 loss: 0.6525\n",
            "Step 640 loss: 0.3171\n",
            "Step 650 loss: 0.8909\n",
            "Step 660 loss: 0.9173\n",
            "Step 670 loss: 0.7708\n",
            "Step 680 loss: 0.5885\n",
            "Step 690 loss: 0.3181\n",
            "Step 700 loss: 0.5215\n",
            "Step 710 loss: 0.4243\n",
            "Step 720 loss: 0.4469\n",
            "Step 730 loss: 0.7125\n",
            "Step 740 loss: 0.6755\n",
            "Step 750 loss: 0.1750\n",
            "Step 760 loss: 0.4782\n",
            "Step 770 loss: 0.6897\n",
            "Step 780 loss: 0.1838\n",
            "Step 790 loss: 0.1927\n",
            "Step 800 loss: 0.5903\n",
            "\n",
            "train_loss: 4.1168 train_rmse: 402.7165 train_mae: 260.3708\n",
            "\n",
            "Step 0 loss: 0.2232\n",
            "Step 10 loss: 0.2366\n",
            "Step 20 loss: 0.2343\n",
            "Step 30 loss: 0.3516\n",
            "Step 40 loss: 0.5329\n",
            "Step 50 loss: 0.2685\n",
            "Step 60 loss: 0.1850\n",
            "Step 70 loss: 0.1952\n",
            "Step 80 loss: 0.6825\n",
            "Step 90 loss: 0.2433\n",
            "Step 100 loss: 0.1476\n",
            "Step 110 loss: 0.3237\n",
            "Step 120 loss: 0.1232\n",
            "Step 130 loss: 0.3306\n",
            "Step 140 loss: 0.3319\n",
            "Step 150 loss: 0.4042\n",
            "Step 160 loss: 0.2437\n",
            "Step 170 loss: 0.2473\n",
            "Step 180 loss: 0.3462\n",
            "Step 190 loss: 0.6347\n",
            "Step 200 loss: 0.1920\n",
            "Step 210 loss: 0.1542\n",
            "Step 220 loss: 0.2827\n",
            "Step 230 loss: 0.2609\n",
            "Step 240 loss: 0.3555\n",
            "Step 250 loss: 0.1654\n",
            "Step 260 loss: 0.1957\n",
            "Step 270 loss: 0.3830\n",
            "Step 280 loss: 0.5984\n",
            "Step 290 loss: 0.3067\n",
            "Step 300 loss: 0.2368\n",
            "Step 310 loss: 0.3552\n",
            "Step 320 loss: 0.3753\n",
            "Step 330 loss: 0.2446\n",
            "Step 340 loss: 0.1753\n",
            "Step 350 loss: 0.4102\n",
            "Step 360 loss: 0.1834\n",
            "Step 370 loss: 0.2987\n",
            "Step 380 loss: 0.2589\n",
            "Step 390 loss: 0.1722\n",
            "Step 400 loss: 0.1562\n",
            "Step 410 loss: 0.7703\n",
            "Step 420 loss: 0.1252\n",
            "Step 430 loss: 0.2791\n",
            "Step 440 loss: 0.2886\n",
            "Step 450 loss: 0.1655\n",
            "Step 460 loss: 0.1715\n",
            "Step 470 loss: 0.2262\n",
            "Step 480 loss: 0.3129\n",
            "Step 490 loss: 0.3341\n",
            "Step 500 loss: 0.6867\n",
            "Step 510 loss: 0.1601\n",
            "Step 520 loss: 0.4689\n",
            "Step 530 loss: 0.2716\n",
            "Step 540 loss: 0.3042\n",
            "Step 550 loss: 0.1588\n",
            "Step 560 loss: 0.2906\n",
            "Step 570 loss: 0.1449\n",
            "Step 580 loss: 0.2227\n",
            "Step 590 loss: 0.1899\n",
            "Step 600 loss: 0.3376\n",
            "Step 610 loss: 0.2743\n",
            "Step 620 loss: 0.1766\n",
            "Step 630 loss: 0.3597\n",
            "Step 640 loss: 0.3288\n",
            "Step 650 loss: 0.1730\n",
            "Step 660 loss: 0.8982\n",
            "Step 670 loss: 0.1867\n",
            "Step 680 loss: 0.2083\n",
            "Step 690 loss: 0.2450\n",
            "Step 700 loss: 0.2416\n",
            "Step 710 loss: 0.2808\n",
            "Step 720 loss: 0.2076\n",
            "Step 730 loss: 0.3167\n",
            "Step 740 loss: 0.2415\n",
            "Step 750 loss: 0.3328\n",
            "Step 760 loss: 0.1810\n",
            "Step 770 loss: 0.3791\n",
            "Step 780 loss: 0.2013\n",
            "Step 790 loss: 0.2743\n",
            "Step 800 loss: 0.2348\n",
            "\n",
            "\n",
            "train_loss: 4.1168 train_rmse: 402.7165 train_mae: 260.3708\n",
            "val_loss: 0.2917 val_rmse: 281.6031 val_mae: 160.1066\n",
            "\n",
            "\n",
            "Epoch 2/10\n",
            "\n",
            "Step 0 loss: 0.2814\n",
            "Step 10 loss: 0.2705\n",
            "Step 20 loss: 0.2521\n",
            "Step 30 loss: 0.2893\n",
            "Step 40 loss: 0.2084\n",
            "Step 50 loss: 0.5200\n",
            "Step 60 loss: 0.2616\n",
            "Step 70 loss: 0.3025\n",
            "Step 80 loss: 0.2007\n",
            "Step 90 loss: 0.2220\n",
            "Step 100 loss: 0.2348\n",
            "Step 110 loss: 0.2183\n",
            "Step 120 loss: 0.1756\n",
            "Step 130 loss: 0.3059\n",
            "Step 140 loss: 0.4089\n",
            "Step 150 loss: 0.1741\n",
            "Step 160 loss: 0.2190\n",
            "Step 170 loss: 0.3492\n",
            "Step 180 loss: 0.2121\n",
            "Step 190 loss: 0.2031\n",
            "Step 200 loss: 0.2349\n",
            "Step 210 loss: 0.1528\n",
            "Step 220 loss: 0.1978\n",
            "Step 230 loss: 0.2908\n",
            "Step 240 loss: 0.2541\n",
            "Step 250 loss: 0.1878\n",
            "Step 260 loss: 0.2540\n",
            "Step 270 loss: 0.2253\n",
            "Step 280 loss: 0.1145\n",
            "Step 290 loss: 0.3217\n",
            "Step 300 loss: 0.1765\n",
            "Step 310 loss: 0.2089\n",
            "Step 320 loss: 0.6796\n",
            "Step 330 loss: 0.1540\n",
            "Step 340 loss: 0.1597\n",
            "Step 350 loss: 0.2421\n",
            "Step 360 loss: 0.1812\n",
            "Step 370 loss: 0.5215\n",
            "Step 380 loss: 0.2494\n",
            "Step 390 loss: 0.2480\n",
            "Step 400 loss: 0.2818\n",
            "Step 410 loss: 0.2977\n",
            "Step 420 loss: 0.4254\n",
            "Step 430 loss: 0.2148\n",
            "Step 440 loss: 0.3022\n",
            "Step 450 loss: 0.3123\n",
            "Step 460 loss: 0.2170\n",
            "Step 470 loss: 0.2393\n",
            "Step 480 loss: 0.1379\n",
            "Step 490 loss: 0.1995\n",
            "Step 500 loss: 0.2881\n",
            "Step 510 loss: 0.1284\n",
            "Step 520 loss: 0.1703\n",
            "Step 530 loss: 0.1948\n",
            "Step 540 loss: 0.1479\n",
            "Step 550 loss: 0.2633\n",
            "Step 560 loss: 0.1307\n",
            "Step 570 loss: 0.1679\n",
            "Step 580 loss: 0.1756\n",
            "Step 590 loss: 0.2872\n",
            "Step 600 loss: 0.1889\n",
            "Step 610 loss: 0.4015\n",
            "Step 620 loss: 0.2071\n",
            "Step 630 loss: 0.1672\n",
            "Step 640 loss: 0.0952\n",
            "Step 650 loss: 0.2172\n",
            "Step 660 loss: 0.1644\n",
            "Step 670 loss: 0.2558\n",
            "Step 680 loss: 0.2315\n",
            "Step 690 loss: 0.0626\n",
            "Step 700 loss: 0.1147\n",
            "Step 710 loss: 0.1566\n",
            "Step 720 loss: 0.1542\n",
            "Step 730 loss: 0.4626\n",
            "Step 740 loss: 0.1268\n",
            "Step 750 loss: 0.1512\n",
            "Step 760 loss: 0.1176\n",
            "Step 770 loss: 0.1677\n",
            "Step 780 loss: 0.1443\n",
            "Step 790 loss: 0.1723\n",
            "Step 800 loss: 0.1751\n",
            "\n",
            "train_loss: 0.2342 train_rmse: 254.6100 train_mae: 144.9800\n",
            "\n",
            "Step 0 loss: 0.0882\n",
            "Step 10 loss: 0.1228\n",
            "Step 20 loss: 0.0505\n",
            "Step 30 loss: 0.1145\n",
            "Step 40 loss: 0.1020\n",
            "Step 50 loss: 0.5315\n",
            "Step 60 loss: 0.0706\n",
            "Step 70 loss: 0.1179\n",
            "Step 80 loss: 0.1188\n",
            "Step 90 loss: 0.1265\n",
            "Step 100 loss: 0.3007\n",
            "Step 110 loss: 0.2803\n",
            "Step 120 loss: 0.2197\n",
            "Step 130 loss: 0.1099\n",
            "Step 140 loss: 0.0897\n",
            "Step 150 loss: 0.1717\n",
            "Step 160 loss: 0.1040\n",
            "Step 170 loss: 0.1749\n",
            "Step 180 loss: 0.1025\n",
            "Step 190 loss: 0.0533\n",
            "Step 200 loss: 0.3970\n",
            "Step 210 loss: 0.1314\n",
            "Step 220 loss: 0.1635\n",
            "Step 230 loss: 0.1133\n",
            "Step 240 loss: 0.1391\n",
            "Step 250 loss: 0.0874\n",
            "Step 260 loss: 0.1288\n",
            "Step 270 loss: 0.0787\n",
            "Step 280 loss: 0.0757\n",
            "Step 290 loss: 0.1118\n",
            "Step 300 loss: 0.1256\n",
            "Step 310 loss: 0.1251\n",
            "Step 320 loss: 0.1696\n",
            "Step 330 loss: 0.0868\n",
            "Step 340 loss: 0.0883\n",
            "Step 350 loss: 0.0850\n",
            "Step 360 loss: 0.0685\n",
            "Step 370 loss: 0.3070\n",
            "Step 380 loss: 0.1444\n",
            "Step 390 loss: 0.1016\n",
            "Step 400 loss: 0.1605\n",
            "Step 410 loss: 0.1549\n",
            "Step 420 loss: 0.2705\n",
            "Step 430 loss: 0.1196\n",
            "Step 440 loss: 0.1028\n",
            "Step 450 loss: 0.1378\n",
            "Step 460 loss: 0.0943\n",
            "Step 470 loss: 0.1234\n",
            "Step 480 loss: 0.1179\n",
            "Step 490 loss: 0.3439\n",
            "Step 500 loss: 0.2813\n",
            "Step 510 loss: 0.1403\n",
            "Step 520 loss: 0.0973\n",
            "Step 530 loss: 0.1848\n",
            "Step 540 loss: 0.1054\n",
            "Step 550 loss: 0.1830\n",
            "Step 560 loss: 0.0790\n",
            "Step 570 loss: 0.0871\n",
            "Step 580 loss: 0.1005\n",
            "Step 590 loss: 0.1005\n",
            "Step 600 loss: 0.1790\n",
            "Step 610 loss: 0.0688\n",
            "Step 620 loss: 0.1197\n",
            "Step 630 loss: 0.3289\n",
            "Step 640 loss: 0.1215\n",
            "Step 650 loss: 0.1713\n",
            "Step 660 loss: 0.0722\n",
            "Step 670 loss: 0.1390\n",
            "Step 680 loss: 0.1361\n",
            "Step 690 loss: 0.1793\n",
            "Step 700 loss: 0.0648\n",
            "Step 710 loss: 0.0865\n",
            "Step 720 loss: 0.1593\n",
            "Step 730 loss: 0.2138\n",
            "Step 740 loss: 0.0968\n",
            "Step 750 loss: 0.2165\n",
            "Step 760 loss: 0.1527\n",
            "Step 770 loss: 0.0998\n",
            "Step 780 loss: 0.2262\n",
            "Step 790 loss: 0.1428\n",
            "Step 800 loss: 0.0686\n",
            "\n",
            "\n",
            "train_loss: 0.2342 train_rmse: 254.6100 train_mae: 144.9800\n",
            "val_loss: 0.1385 val_rmse: 193.7036 val_mae: 105.3862\n",
            "\n",
            "\n",
            "Epoch 3/10\n",
            "\n",
            "Step 0 loss: 0.1683\n",
            "Step 10 loss: 0.1349\n",
            "Step 20 loss: 0.1821\n",
            "Step 30 loss: 0.1030\n",
            "Step 40 loss: 0.3774\n",
            "Step 50 loss: 0.1217\n",
            "Step 60 loss: 0.1203\n",
            "Step 70 loss: 0.3127\n",
            "Step 80 loss: 0.1434\n",
            "Step 90 loss: 0.1894\n",
            "Step 100 loss: 0.1689\n",
            "Step 110 loss: 0.4258\n",
            "Step 120 loss: 0.4475\n",
            "Step 130 loss: 0.1495\n",
            "Step 140 loss: 0.0798\n",
            "Step 150 loss: 0.1970\n",
            "Step 160 loss: 0.2324\n",
            "Step 170 loss: 0.0656\n",
            "Step 180 loss: 0.0871\n",
            "Step 190 loss: 0.1672\n",
            "Step 200 loss: 0.1991\n",
            "Step 210 loss: 0.1039\n",
            "Step 220 loss: 0.1319\n",
            "Step 230 loss: 0.1813\n",
            "Step 240 loss: 0.1718\n",
            "Step 250 loss: 0.1877\n",
            "Step 260 loss: 0.0814\n",
            "Step 270 loss: 0.1266\n",
            "Step 280 loss: 0.0637\n",
            "Step 290 loss: 0.0895\n",
            "Step 300 loss: 0.1602\n",
            "Step 310 loss: 0.2240\n",
            "Step 320 loss: 0.1743\n",
            "Step 330 loss: 0.3453\n",
            "Step 340 loss: 0.0844\n",
            "Step 350 loss: 0.1914\n",
            "Step 360 loss: 0.2293\n",
            "Step 370 loss: 0.1548\n",
            "Step 380 loss: 0.0951\n",
            "Step 390 loss: 0.0934\n",
            "Step 400 loss: 0.2833\n",
            "Step 410 loss: 0.0747\n",
            "Step 420 loss: 0.1566\n",
            "Step 430 loss: 0.0698\n",
            "Step 440 loss: 0.0949\n",
            "Step 450 loss: 0.1359\n",
            "Step 460 loss: 0.0996\n",
            "Step 470 loss: 0.1045\n",
            "Step 480 loss: 0.2596\n",
            "Step 490 loss: 0.1359\n",
            "Step 500 loss: 0.1284\n",
            "Step 510 loss: 0.1745\n",
            "Step 520 loss: 0.1009\n",
            "Step 530 loss: 0.1600\n",
            "Step 540 loss: 0.1382\n",
            "Step 550 loss: 0.1932\n",
            "Step 560 loss: 0.0841\n",
            "Step 570 loss: 0.1603\n",
            "Step 580 loss: 0.1300\n",
            "Step 590 loss: 0.1592\n",
            "Step 600 loss: 0.1270\n",
            "Step 610 loss: 0.1166\n",
            "Step 620 loss: 0.0944\n",
            "Step 630 loss: 0.2920\n",
            "Step 640 loss: 0.1300\n",
            "Step 650 loss: 0.0908\n",
            "Step 660 loss: 0.1172\n",
            "Step 670 loss: 0.0786\n",
            "Step 680 loss: 0.1543\n",
            "Step 690 loss: 0.1393\n",
            "Step 700 loss: 0.0835\n",
            "Step 710 loss: 0.1056\n",
            "Step 720 loss: 0.1144\n",
            "Step 730 loss: 0.1948\n",
            "Step 740 loss: 0.1103\n",
            "Step 750 loss: 0.0950\n",
            "Step 760 loss: 0.1096\n",
            "Step 770 loss: 0.1948\n",
            "Step 780 loss: 0.0942\n",
            "Step 790 loss: 0.0819\n",
            "Step 800 loss: 0.0750\n",
            "\n",
            "train_loss: 0.1539 train_rmse: 187.7862 train_mae: 110.9232\n",
            "\n",
            "Step 0 loss: 0.0998\n",
            "Step 10 loss: 0.0881\n",
            "Step 20 loss: 0.1629\n",
            "Step 30 loss: 0.0879\n",
            "Step 40 loss: 0.0489\n",
            "Step 50 loss: 0.2140\n",
            "Step 60 loss: 0.0960\n",
            "Step 70 loss: 0.0920\n",
            "Step 80 loss: 0.1211\n",
            "Step 90 loss: 0.0860\n",
            "Step 100 loss: 0.1034\n",
            "Step 110 loss: 0.0983\n",
            "Step 120 loss: 0.0574\n",
            "Step 130 loss: 0.1765\n",
            "Step 140 loss: 0.0623\n",
            "Step 150 loss: 0.0611\n",
            "Step 160 loss: 0.0555\n",
            "Step 170 loss: 0.0365\n",
            "Step 180 loss: 0.1086\n",
            "Step 190 loss: 0.0523\n",
            "Step 200 loss: 0.0525\n",
            "Step 210 loss: 0.1209\n",
            "Step 220 loss: 0.0367\n",
            "Step 230 loss: 0.0418\n",
            "Step 240 loss: 0.0447\n",
            "Step 250 loss: 0.0653\n",
            "Step 260 loss: 0.0665\n",
            "Step 270 loss: 0.0882\n",
            "Step 280 loss: 0.1343\n",
            "Step 290 loss: 0.0609\n",
            "Step 300 loss: 0.4401\n",
            "Step 310 loss: 0.1480\n",
            "Step 320 loss: 0.0671\n",
            "Step 330 loss: 0.0886\n",
            "Step 340 loss: 0.0516\n",
            "Step 350 loss: 0.0517\n",
            "Step 360 loss: 0.0534\n",
            "Step 370 loss: 0.1425\n",
            "Step 380 loss: 0.0356\n",
            "Step 390 loss: 0.0554\n",
            "Step 400 loss: 0.0682\n",
            "Step 410 loss: 0.0920\n",
            "Step 420 loss: 0.0826\n",
            "Step 430 loss: 0.0369\n",
            "Step 440 loss: 0.0660\n",
            "Step 450 loss: 0.2918\n",
            "Step 460 loss: 0.2593\n",
            "Step 470 loss: 0.1014\n",
            "Step 480 loss: 0.0641\n",
            "Step 490 loss: 0.0578\n",
            "Step 500 loss: 0.3006\n",
            "Step 510 loss: 0.0447\n",
            "Step 520 loss: 0.1695\n",
            "Step 530 loss: 0.0518\n",
            "Step 540 loss: 0.1648\n",
            "Step 550 loss: 0.0898\n",
            "Step 560 loss: 0.0735\n",
            "Step 570 loss: 0.2906\n",
            "Step 580 loss: 0.0885\n",
            "Step 590 loss: 0.0900\n",
            "Step 600 loss: 0.0857\n",
            "Step 610 loss: 0.0693\n",
            "Step 620 loss: 0.0597\n",
            "Step 630 loss: 0.0745\n",
            "Step 640 loss: 0.2400\n",
            "Step 650 loss: 0.0613\n",
            "Step 660 loss: 0.2135\n",
            "Step 670 loss: 0.1579\n",
            "Step 680 loss: 0.2350\n",
            "Step 690 loss: 0.0902\n",
            "Step 700 loss: 0.0648\n",
            "Step 710 loss: 0.0862\n",
            "Step 720 loss: 0.0906\n",
            "Step 730 loss: 0.0673\n",
            "Step 740 loss: 0.0818\n",
            "Step 750 loss: 0.1051\n",
            "Step 760 loss: 0.1342\n",
            "Step 770 loss: 0.1035\n",
            "Step 780 loss: 0.0583\n",
            "Step 790 loss: 0.1246\n",
            "Step 800 loss: 0.2848\n",
            "\n",
            "\n",
            "train_loss: 0.1539 train_rmse: 187.7862 train_mae: 110.9232\n",
            "val_loss: 0.1047 val_rmse: 145.0155 val_mae: 84.1515\n",
            "\n",
            "\n",
            "Epoch 4/10\n",
            "\n",
            "Step 0 loss: 0.1080\n",
            "Step 10 loss: 0.0997\n",
            "Step 20 loss: 0.1379\n",
            "Step 30 loss: 0.1034\n",
            "Step 40 loss: 0.1086\n",
            "Step 50 loss: 0.0603\n",
            "Step 60 loss: 0.1712\n",
            "Step 70 loss: 0.0713\n",
            "Step 80 loss: 0.0805\n",
            "Step 90 loss: 0.3805\n",
            "Step 100 loss: 0.1644\n",
            "Step 110 loss: 0.1503\n",
            "Step 120 loss: 0.1485\n",
            "Step 130 loss: 0.1428\n",
            "Step 140 loss: 0.1176\n",
            "Step 150 loss: 0.1382\n",
            "Step 160 loss: 0.1586\n",
            "Step 170 loss: 0.1056\n",
            "Step 180 loss: 0.1337\n",
            "Step 190 loss: 0.2341\n",
            "Step 200 loss: 0.1078\n",
            "Step 210 loss: 0.0845\n",
            "Step 220 loss: 0.2085\n",
            "Step 230 loss: 0.1882\n",
            "Step 240 loss: 0.1823\n",
            "Step 250 loss: 0.1167\n",
            "Step 260 loss: 0.1330\n",
            "Step 270 loss: 0.1048\n",
            "Step 280 loss: 0.1983\n",
            "Step 290 loss: 0.1805\n",
            "Step 300 loss: 0.2114\n",
            "Step 310 loss: 0.1156\n",
            "Step 320 loss: 0.1758\n",
            "Step 330 loss: 0.0655\n",
            "Step 340 loss: 0.3400\n",
            "Step 350 loss: 0.1465\n",
            "Step 360 loss: 0.2552\n",
            "Step 370 loss: 0.5153\n",
            "Step 380 loss: 0.1021\n",
            "Step 390 loss: 0.1225\n",
            "Step 400 loss: 0.1051\n",
            "Step 410 loss: 0.1456\n",
            "Step 420 loss: 0.0678\n",
            "Step 430 loss: 0.0949\n",
            "Step 440 loss: 0.0733\n",
            "Step 450 loss: 0.0759\n",
            "Step 460 loss: 0.0948\n",
            "Step 470 loss: 0.1166\n",
            "Step 480 loss: 0.1612\n",
            "Step 490 loss: 0.1269\n",
            "Step 500 loss: 0.1206\n",
            "Step 510 loss: 0.1090\n",
            "Step 520 loss: 0.0753\n",
            "Step 530 loss: 0.0782\n",
            "Step 540 loss: 0.0880\n",
            "Step 550 loss: 0.0910\n",
            "Step 560 loss: 0.0914\n",
            "Step 570 loss: 0.0565\n",
            "Step 580 loss: 0.0841\n",
            "Step 590 loss: 0.1798\n",
            "Step 600 loss: 0.0686\n",
            "Step 610 loss: 0.1014\n",
            "Step 620 loss: 0.0829\n",
            "Step 630 loss: 0.0853\n",
            "Step 640 loss: 0.0751\n",
            "Step 650 loss: 0.1707\n",
            "Step 660 loss: 0.0718\n",
            "Step 670 loss: 0.1216\n",
            "Step 680 loss: 0.2345\n",
            "Step 690 loss: 0.2228\n",
            "Step 700 loss: 0.0815\n",
            "Step 710 loss: 0.1068\n",
            "Step 720 loss: 0.0920\n",
            "Step 730 loss: 0.0733\n",
            "Step 740 loss: 0.0983\n",
            "Step 750 loss: 0.0757\n",
            "Step 760 loss: 0.1473\n",
            "Step 770 loss: 0.2665\n",
            "Step 780 loss: 0.0916\n",
            "Step 790 loss: 0.1028\n",
            "Step 800 loss: 0.0568\n",
            "\n",
            "train_loss: 0.1314 train_rmse: 162.4616 train_mae: 98.4482\n",
            "\n",
            "Step 0 loss: 0.0522\n",
            "Step 10 loss: 0.0513\n",
            "Step 20 loss: 0.1516\n",
            "Step 30 loss: 0.1905\n",
            "Step 40 loss: 0.0353\n",
            "Step 50 loss: 0.0676\n",
            "Step 60 loss: 0.1263\n",
            "Step 70 loss: 0.1044\n",
            "Step 80 loss: 0.0950\n",
            "Step 90 loss: 0.0670\n",
            "Step 100 loss: 0.0556\n",
            "Step 110 loss: 0.0982\n",
            "Step 120 loss: 0.0733\n",
            "Step 130 loss: 0.1868\n",
            "Step 140 loss: 0.0520\n",
            "Step 150 loss: 0.0576\n",
            "Step 160 loss: 0.1665\n",
            "Step 170 loss: 0.2089\n",
            "Step 180 loss: 0.1002\n",
            "Step 190 loss: 0.0385\n",
            "Step 200 loss: 0.1182\n",
            "Step 210 loss: 0.0554\n",
            "Step 220 loss: 0.0509\n",
            "Step 230 loss: 0.1156\n",
            "Step 240 loss: 0.0717\n",
            "Step 250 loss: 0.0804\n",
            "Step 260 loss: 0.1141\n",
            "Step 270 loss: 0.1136\n",
            "Step 280 loss: 0.0477\n",
            "Step 290 loss: 0.0806\n",
            "Step 300 loss: 0.0944\n",
            "Step 310 loss: 0.1056\n",
            "Step 320 loss: 0.0798\n",
            "Step 330 loss: 0.0895\n",
            "Step 340 loss: 0.0761\n",
            "Step 350 loss: 0.0672\n",
            "Step 360 loss: 0.1283\n",
            "Step 370 loss: 0.0517\n",
            "Step 380 loss: 0.0480\n",
            "Step 390 loss: 0.0955\n",
            "Step 400 loss: 0.1112\n",
            "Step 410 loss: 0.1920\n",
            "Step 420 loss: 0.0438\n",
            "Step 430 loss: 0.1777\n",
            "Step 440 loss: 0.0875\n",
            "Step 450 loss: 0.1477\n",
            "Step 460 loss: 0.0264\n",
            "Step 470 loss: 0.0740\n",
            "Step 480 loss: 0.0776\n",
            "Step 490 loss: 0.1327\n",
            "Step 500 loss: 0.0589\n",
            "Step 510 loss: 0.0369\n",
            "Step 520 loss: 0.0920\n",
            "Step 530 loss: 0.0362\n",
            "Step 540 loss: 0.1077\n",
            "Step 550 loss: 0.0654\n",
            "Step 560 loss: 0.0411\n",
            "Step 570 loss: 0.0466\n",
            "Step 580 loss: 0.1507\n",
            "Step 590 loss: 0.2757\n",
            "Step 600 loss: 0.0796\n",
            "Step 610 loss: 0.0607\n",
            "Step 620 loss: 0.0618\n",
            "Step 630 loss: 0.0505\n",
            "Step 640 loss: 0.0497\n",
            "Step 650 loss: 0.0981\n",
            "Step 660 loss: 0.0452\n",
            "Step 670 loss: 0.0692\n",
            "Step 680 loss: 0.0422\n",
            "Step 690 loss: 0.0615\n",
            "Step 700 loss: 0.0818\n",
            "Step 710 loss: 0.0960\n",
            "Step 720 loss: 0.1376\n",
            "Step 730 loss: 0.0433\n",
            "Step 740 loss: 0.0699\n",
            "Step 750 loss: 0.0916\n",
            "Step 760 loss: 0.0446\n",
            "Step 770 loss: 0.0842\n",
            "Step 780 loss: 0.0358\n",
            "Step 790 loss: 0.0981\n",
            "Step 800 loss: 0.0950\n",
            "\n",
            "\n",
            "train_loss: 0.1314 train_rmse: 162.4616 train_mae: 98.4482\n",
            "val_loss: 0.0934 val_rmse: 117.2461 val_mae: 71.0008\n",
            "\n",
            "\n",
            "Epoch 5/10\n",
            "\n",
            "Step 0 loss: 0.1121\n",
            "Step 10 loss: 0.1099\n",
            "Step 20 loss: 0.1157\n",
            "Step 30 loss: 0.1273\n",
            "Step 40 loss: 0.0335\n",
            "Step 50 loss: 0.0824\n",
            "Step 60 loss: 0.0920\n",
            "Step 70 loss: 0.1267\n",
            "Step 80 loss: 0.0909\n",
            "Step 90 loss: 0.1564\n",
            "Step 100 loss: 0.1803\n",
            "Step 110 loss: 0.1124\n",
            "Step 120 loss: 0.1128\n",
            "Step 130 loss: 0.1240\n",
            "Step 140 loss: 0.0776\n",
            "Step 150 loss: 0.0789\n",
            "Step 160 loss: 0.1957\n",
            "Step 170 loss: 0.0676\n",
            "Step 180 loss: 0.0936\n",
            "Step 190 loss: 0.1554\n",
            "Step 200 loss: 0.0655\n",
            "Step 210 loss: 0.0568\n",
            "Step 220 loss: 0.0555\n",
            "Step 230 loss: 0.2810\n",
            "Step 240 loss: 0.0845\n",
            "Step 250 loss: 0.1210\n",
            "Step 260 loss: 0.1511\n",
            "Step 270 loss: 0.0807\n",
            "Step 280 loss: 0.0927\n",
            "Step 290 loss: 0.2130\n",
            "Step 300 loss: 0.1500\n",
            "Step 310 loss: 0.0605\n",
            "Step 320 loss: 0.1064\n",
            "Step 330 loss: 0.2046\n",
            "Step 340 loss: 0.0744\n",
            "Step 350 loss: 0.1099\n",
            "Step 360 loss: 0.0571\n",
            "Step 370 loss: 0.0652\n",
            "Step 380 loss: 0.0877\n",
            "Step 390 loss: 0.1342\n",
            "Step 400 loss: 0.1351\n",
            "Step 410 loss: 0.1472\n",
            "Step 420 loss: 0.1306\n",
            "Step 430 loss: 0.1471\n",
            "Step 440 loss: 0.0807\n",
            "Step 450 loss: 0.0804\n",
            "Step 460 loss: 0.0334\n",
            "Step 470 loss: 0.4040\n",
            "Step 480 loss: 0.0576\n",
            "Step 490 loss: 0.0869\n",
            "Step 500 loss: 0.1390\n",
            "Step 510 loss: 0.0884\n",
            "Step 520 loss: 0.0839\n",
            "Step 530 loss: 0.0753\n",
            "Step 540 loss: 0.1016\n",
            "Step 550 loss: 0.1729\n",
            "Step 560 loss: 0.0881\n",
            "Step 570 loss: 0.0776\n",
            "Step 580 loss: 0.1068\n",
            "Step 590 loss: 0.2153\n",
            "Step 600 loss: 0.2153\n",
            "Step 610 loss: 0.0843\n",
            "Step 620 loss: 0.2237\n",
            "Step 630 loss: 0.1050\n",
            "Step 640 loss: 0.1527\n",
            "Step 650 loss: 0.1125\n",
            "Step 660 loss: 0.0977\n",
            "Step 670 loss: 0.2441\n",
            "Step 680 loss: 0.0608\n",
            "Step 690 loss: 0.1555\n",
            "Step 700 loss: 0.1066\n",
            "Step 710 loss: 0.1437\n",
            "Step 720 loss: 0.1572\n",
            "Step 730 loss: 0.1757\n",
            "Step 740 loss: 0.1229\n",
            "Step 750 loss: 0.1708\n",
            "Step 760 loss: 0.0789\n",
            "Step 770 loss: 0.0814\n",
            "Step 780 loss: 0.0930\n",
            "Step 790 loss: 0.0893\n",
            "Step 800 loss: 0.0621\n",
            "\n",
            "train_loss: 0.1217 train_rmse: 152.4583 train_mae: 93.6991\n",
            "\n",
            "Step 0 loss: 0.0866\n",
            "Step 10 loss: 0.0442\n",
            "Step 20 loss: 0.0401\n",
            "Step 30 loss: 0.0440\n",
            "Step 40 loss: 0.0495\n",
            "Step 50 loss: 0.0361\n",
            "Step 60 loss: 0.0265\n",
            "Step 70 loss: 0.0470\n",
            "Step 80 loss: 0.0506\n",
            "Step 90 loss: 0.0579\n",
            "Step 100 loss: 0.0769\n",
            "Step 110 loss: 0.0523\n",
            "Step 120 loss: 0.0525\n",
            "Step 130 loss: 0.1597\n",
            "Step 140 loss: 0.0374\n",
            "Step 150 loss: 0.0426\n",
            "Step 160 loss: 0.0316\n",
            "Step 170 loss: 0.0478\n",
            "Step 180 loss: 0.0756\n",
            "Step 190 loss: 0.0613\n",
            "Step 200 loss: 0.2236\n",
            "Step 210 loss: 0.0311\n",
            "Step 220 loss: 0.0474\n",
            "Step 230 loss: 0.0247\n",
            "Step 240 loss: 0.1461\n",
            "Step 250 loss: 0.1656\n",
            "Step 260 loss: 0.0422\n",
            "Step 270 loss: 0.1119\n",
            "Step 280 loss: 0.0672\n",
            "Step 290 loss: 0.0324\n",
            "Step 300 loss: 0.1041\n",
            "Step 310 loss: 0.0354\n",
            "Step 320 loss: 0.0577\n",
            "Step 330 loss: 0.1421\n",
            "Step 340 loss: 0.1720\n",
            "Step 350 loss: 0.2150\n",
            "Step 360 loss: 0.0547\n",
            "Step 370 loss: 0.0297\n",
            "Step 380 loss: 0.0662\n",
            "Step 390 loss: 0.2440\n",
            "Step 400 loss: 0.0404\n",
            "Step 410 loss: 0.0426\n",
            "Step 420 loss: 0.1652\n",
            "Step 430 loss: 0.3707\n",
            "Step 440 loss: 0.1348\n",
            "Step 450 loss: 0.3571\n",
            "Step 460 loss: 0.2750\n",
            "Step 470 loss: 0.0297\n",
            "Step 480 loss: 0.0309\n",
            "Step 490 loss: 0.0558\n",
            "Step 500 loss: 0.0651\n",
            "Step 510 loss: 0.1304\n",
            "Step 520 loss: 0.1062\n",
            "Step 530 loss: 0.0860\n",
            "Step 540 loss: 0.2576\n",
            "Step 550 loss: 0.0796\n",
            "Step 560 loss: 0.0550\n",
            "Step 570 loss: 0.0858\n",
            "Step 580 loss: 0.0543\n",
            "Step 590 loss: 0.0641\n",
            "Step 600 loss: 0.0350\n",
            "Step 610 loss: 0.0438\n",
            "Step 620 loss: 0.0283\n",
            "Step 630 loss: 0.0583\n",
            "Step 640 loss: 0.0399\n",
            "Step 650 loss: 0.1374\n",
            "Step 660 loss: 0.1128\n",
            "Step 670 loss: 0.0490\n",
            "Step 680 loss: 0.0804\n",
            "Step 690 loss: 0.0651\n",
            "Step 700 loss: 0.0585\n",
            "Step 710 loss: 0.0939\n",
            "Step 720 loss: 0.0511\n",
            "Step 730 loss: 0.0462\n",
            "Step 740 loss: 0.1251\n",
            "Step 750 loss: 0.0980\n",
            "Step 760 loss: 0.0961\n",
            "Step 770 loss: 0.0234\n",
            "Step 780 loss: 0.0405\n",
            "Step 790 loss: 0.0730\n",
            "Step 800 loss: 0.0304\n",
            "\n",
            "\n",
            "train_loss: 0.1217 train_rmse: 152.4583 train_mae: 93.6991\n",
            "val_loss: 0.0847 val_rmse: 107.2697 val_mae: 65.8276\n",
            "\n",
            "\n",
            "Epoch 6/10\n",
            "\n",
            "Step 0 loss: 0.2369\n",
            "Step 10 loss: 0.2786\n",
            "Step 20 loss: 0.2561\n",
            "Step 30 loss: 0.1463\n",
            "Step 40 loss: 0.0675\n",
            "Step 50 loss: 0.1211\n",
            "Step 60 loss: 0.2629\n",
            "Step 70 loss: 0.0618\n",
            "Step 80 loss: 0.0535\n",
            "Step 90 loss: 0.0695\n",
            "Step 100 loss: 0.1099\n",
            "Step 110 loss: 0.0830\n",
            "Step 120 loss: 0.1452\n",
            "Step 130 loss: 0.2291\n",
            "Step 140 loss: 0.1289\n",
            "Step 150 loss: 0.1830\n",
            "Step 160 loss: 0.0687\n",
            "Step 170 loss: 0.1507\n",
            "Step 180 loss: 0.0520\n",
            "Step 190 loss: 0.2641\n",
            "Step 200 loss: 0.1805\n",
            "Step 210 loss: 0.2970\n",
            "Step 220 loss: 0.0576\n",
            "Step 230 loss: 0.1883\n",
            "Step 240 loss: 0.0734\n",
            "Step 250 loss: 0.0834\n",
            "Step 260 loss: 0.0631\n",
            "Step 270 loss: 0.0851\n",
            "Step 280 loss: 0.1815\n",
            "Step 290 loss: 0.1206\n",
            "Step 300 loss: 0.1398\n",
            "Step 310 loss: 0.1803\n",
            "Step 320 loss: 0.1031\n",
            "Step 330 loss: 0.1101\n",
            "Step 340 loss: 0.1106\n",
            "Step 350 loss: 0.0733\n",
            "Step 360 loss: 0.0748\n",
            "Step 370 loss: 0.0759\n",
            "Step 380 loss: 0.1508\n",
            "Step 390 loss: 0.0558\n",
            "Step 400 loss: 0.1316\n",
            "Step 410 loss: 0.0679\n",
            "Step 420 loss: 0.2341\n",
            "Step 430 loss: 0.0903\n",
            "Step 440 loss: 0.1178\n",
            "Step 450 loss: 0.0680\n",
            "Step 460 loss: 0.1117\n",
            "Step 470 loss: 0.1043\n",
            "Step 480 loss: 0.1760\n",
            "Step 490 loss: 0.0767\n",
            "Step 500 loss: 0.0814\n",
            "Step 510 loss: 0.0787\n",
            "Step 520 loss: 0.1030\n",
            "Step 530 loss: 0.1201\n",
            "Step 540 loss: 0.0693\n",
            "Step 550 loss: 0.0671\n",
            "Step 560 loss: 0.0787\n",
            "Step 570 loss: 0.1008\n",
            "Step 580 loss: 0.0903\n",
            "Step 590 loss: 0.1377\n",
            "Step 600 loss: 0.0717\n",
            "Step 610 loss: 0.1610\n",
            "Step 620 loss: 0.0733\n",
            "Step 630 loss: 0.0966\n",
            "Step 640 loss: 0.1363\n",
            "Step 650 loss: 0.0898\n",
            "Step 660 loss: 0.1394\n",
            "Step 670 loss: 0.1664\n",
            "Step 680 loss: 0.0589\n",
            "Step 690 loss: 0.0783\n",
            "Step 700 loss: 0.1267\n",
            "Step 710 loss: 0.1952\n",
            "Step 720 loss: 0.1636\n",
            "Step 730 loss: 0.1925\n",
            "Step 740 loss: 0.2520\n",
            "Step 750 loss: 0.0538\n",
            "Step 760 loss: 0.1034\n",
            "Step 770 loss: 0.5768\n",
            "Step 780 loss: 0.0597\n",
            "Step 790 loss: 0.1269\n",
            "Step 800 loss: 0.0694\n",
            "\n",
            "train_loss: 0.1164 train_rmse: 148.9115 train_mae: 91.2382\n",
            "\n",
            "Step 0 loss: 0.0676\n",
            "Step 10 loss: 0.0876\n",
            "Step 20 loss: 0.1237\n",
            "Step 30 loss: 0.0684\n",
            "Step 40 loss: 0.0957\n",
            "Step 50 loss: 0.1307\n",
            "Step 60 loss: 0.0310\n",
            "Step 70 loss: 0.0579\n",
            "Step 80 loss: 0.0270\n",
            "Step 90 loss: 0.0241\n",
            "Step 100 loss: 0.0656\n",
            "Step 110 loss: 0.1683\n",
            "Step 120 loss: 0.0808\n",
            "Step 130 loss: 0.1322\n",
            "Step 140 loss: 0.0646\n",
            "Step 150 loss: 0.1705\n",
            "Step 160 loss: 0.0529\n",
            "Step 170 loss: 0.0369\n",
            "Step 180 loss: 0.0494\n",
            "Step 190 loss: 0.2241\n",
            "Step 200 loss: 0.0626\n",
            "Step 210 loss: 0.0556\n",
            "Step 220 loss: 0.0426\n",
            "Step 230 loss: 0.0490\n",
            "Step 240 loss: 0.0554\n",
            "Step 250 loss: 0.1752\n",
            "Step 260 loss: 0.1034\n",
            "Step 270 loss: 0.0312\n",
            "Step 280 loss: 0.0309\n",
            "Step 290 loss: 0.0228\n",
            "Step 300 loss: 0.0310\n",
            "Step 310 loss: 0.2877\n",
            "Step 320 loss: 0.0433\n",
            "Step 330 loss: 0.0624\n",
            "Step 340 loss: 0.1554\n",
            "Step 350 loss: 0.0462\n",
            "Step 360 loss: 0.0417\n",
            "Step 370 loss: 0.1593\n",
            "Step 380 loss: 0.0580\n",
            "Step 390 loss: 0.0307\n",
            "Step 400 loss: 0.0488\n",
            "Step 410 loss: 0.0403\n",
            "Step 420 loss: 0.2735\n",
            "Step 430 loss: 0.0433\n",
            "Step 440 loss: 0.0600\n",
            "Step 450 loss: 0.0704\n",
            "Step 460 loss: 0.0547\n",
            "Step 470 loss: 0.0592\n",
            "Step 480 loss: 0.0671\n",
            "Step 490 loss: 0.0280\n",
            "Step 500 loss: 0.0269\n",
            "Step 510 loss: 0.1335\n",
            "Step 520 loss: 0.1058\n",
            "Step 530 loss: 0.0483\n",
            "Step 540 loss: 0.0459\n",
            "Step 550 loss: 0.1565\n",
            "Step 560 loss: 0.0471\n",
            "Step 570 loss: 0.0392\n",
            "Step 580 loss: 0.0402\n",
            "Step 590 loss: 0.0234\n",
            "Step 600 loss: 0.0905\n",
            "Step 610 loss: 0.0823\n",
            "Step 620 loss: 0.0745\n",
            "Step 630 loss: 0.0956\n",
            "Step 640 loss: 0.1925\n",
            "Step 650 loss: 0.0304\n",
            "Step 660 loss: 0.0646\n",
            "Step 670 loss: 0.0540\n",
            "Step 680 loss: 0.0342\n",
            "Step 690 loss: 0.0716\n",
            "Step 700 loss: 0.0423\n",
            "Step 710 loss: 0.0502\n",
            "Step 720 loss: 0.0458\n",
            "Step 730 loss: 0.3480\n",
            "Step 740 loss: 0.0630\n",
            "Step 750 loss: 0.0218\n",
            "Step 760 loss: 0.0318\n",
            "Step 770 loss: 0.0442\n",
            "Step 780 loss: 0.1042\n",
            "Step 790 loss: 0.0723\n",
            "Step 800 loss: 0.1540\n",
            "\n",
            "\n",
            "train_loss: 0.1164 train_rmse: 148.9115 train_mae: 91.2382\n",
            "val_loss: 0.0830 val_rmse: 121.7504 val_mae: 70.2506\n",
            "\n",
            "\n",
            "Epoch 7/10\n",
            "\n",
            "Step 0 loss: 0.0829\n",
            "Step 10 loss: 0.1900\n",
            "Step 20 loss: 0.0538\n",
            "Step 30 loss: 0.0844\n",
            "Step 40 loss: 0.0497\n",
            "Step 50 loss: 0.0655\n",
            "Step 60 loss: 0.0602\n",
            "Step 70 loss: 0.0848\n",
            "Step 80 loss: 0.1957\n",
            "Step 90 loss: 0.1092\n",
            "Step 100 loss: 0.1183\n",
            "Step 110 loss: 0.1764\n",
            "Step 120 loss: 0.1732\n",
            "Step 130 loss: 0.1202\n",
            "Step 140 loss: 0.0726\n",
            "Step 150 loss: 0.0695\n",
            "Step 160 loss: 0.0716\n",
            "Step 170 loss: 0.0540\n",
            "Step 180 loss: 0.0814\n",
            "Step 190 loss: 0.1320\n",
            "Step 200 loss: 0.0285\n",
            "Step 210 loss: 0.0846\n",
            "Step 220 loss: 0.0961\n",
            "Step 230 loss: 0.1326\n",
            "Step 240 loss: 0.1354\n",
            "Step 250 loss: 0.1383\n",
            "Step 260 loss: 0.1062\n",
            "Step 270 loss: 0.0772\n",
            "Step 280 loss: 0.1591\n",
            "Step 290 loss: 0.0975\n",
            "Step 300 loss: 0.0867\n",
            "Step 310 loss: 0.0859\n",
            "Step 320 loss: 0.0985\n",
            "Step 330 loss: 0.1196\n",
            "Step 340 loss: 0.1050\n",
            "Step 350 loss: 0.3056\n",
            "Step 360 loss: 0.0922\n",
            "Step 370 loss: 0.0888\n",
            "Step 380 loss: 0.0714\n",
            "Step 390 loss: 0.0893\n",
            "Step 400 loss: 0.0745\n",
            "Step 410 loss: 0.0844\n",
            "Step 420 loss: 0.1414\n",
            "Step 430 loss: 0.2778\n",
            "Step 440 loss: 0.0406\n",
            "Step 450 loss: 0.0326\n",
            "Step 460 loss: 0.1345\n",
            "Step 470 loss: 0.1051\n",
            "Step 480 loss: 0.1042\n",
            "Step 490 loss: 0.0851\n",
            "Step 500 loss: 0.1334\n",
            "Step 510 loss: 0.0663\n",
            "Step 520 loss: 0.0710\n",
            "Step 530 loss: 0.0794\n",
            "Step 540 loss: 0.1550\n",
            "Step 550 loss: 0.0730\n",
            "Step 560 loss: 0.0602\n",
            "Step 570 loss: 0.0426\n",
            "Step 580 loss: 0.0847\n",
            "Step 590 loss: 0.0948\n",
            "Step 600 loss: 0.0877\n",
            "Step 610 loss: 0.0303\n",
            "Step 620 loss: 0.0559\n",
            "Step 630 loss: 0.0898\n",
            "Step 640 loss: 0.0597\n",
            "Step 650 loss: 0.2015\n",
            "Step 660 loss: 0.0511\n",
            "Step 670 loss: 0.1566\n",
            "Step 680 loss: 0.0752\n",
            "Step 690 loss: 0.0854\n",
            "Step 700 loss: 0.0496\n",
            "Step 710 loss: 0.1598\n",
            "Step 720 loss: 0.2213\n",
            "Step 730 loss: 0.0913\n",
            "Step 740 loss: 0.0399\n",
            "Step 750 loss: 0.0870\n",
            "Step 760 loss: 0.1301\n",
            "Step 770 loss: 0.0936\n",
            "Step 780 loss: 0.1936\n",
            "Step 790 loss: 0.2646\n",
            "Step 800 loss: 0.1799\n",
            "\n",
            "train_loss: 0.1081 train_rmse: 142.1004 train_mae: 87.5396\n",
            "\n",
            "Step 0 loss: 0.0648\n",
            "Step 10 loss: 0.0614\n",
            "Step 20 loss: 0.1442\n",
            "Step 30 loss: 0.0630\n",
            "Step 40 loss: 0.0785\n",
            "Step 50 loss: 0.0919\n",
            "Step 60 loss: 0.0464\n",
            "Step 70 loss: 0.0332\n",
            "Step 80 loss: 0.0743\n",
            "Step 90 loss: 0.0532\n",
            "Step 100 loss: 0.0474\n",
            "Step 110 loss: 0.0700\n",
            "Step 120 loss: 0.0257\n",
            "Step 130 loss: 0.0283\n",
            "Step 140 loss: 0.0559\n",
            "Step 150 loss: 0.0412\n",
            "Step 160 loss: 0.0525\n",
            "Step 170 loss: 0.1033\n",
            "Step 180 loss: 0.0952\n",
            "Step 190 loss: 0.1258\n",
            "Step 200 loss: 0.0681\n",
            "Step 210 loss: 0.0667\n",
            "Step 220 loss: 0.0540\n",
            "Step 230 loss: 0.0376\n",
            "Step 240 loss: 0.0428\n",
            "Step 250 loss: 0.2018\n",
            "Step 260 loss: 0.0610\n",
            "Step 270 loss: 0.0805\n",
            "Step 280 loss: 0.0642\n",
            "Step 290 loss: 0.3644\n",
            "Step 300 loss: 0.0433\n",
            "Step 310 loss: 0.0340\n",
            "Step 320 loss: 0.0232\n",
            "Step 330 loss: 0.0301\n",
            "Step 340 loss: 0.0285\n",
            "Step 350 loss: 0.0225\n",
            "Step 360 loss: 0.0360\n",
            "Step 370 loss: 0.0699\n",
            "Step 380 loss: 0.0563\n",
            "Step 390 loss: 0.0447\n",
            "Step 400 loss: 0.0566\n",
            "Step 410 loss: 0.0830\n",
            "Step 420 loss: 0.0815\n",
            "Step 430 loss: 0.0908\n",
            "Step 440 loss: 0.1004\n",
            "Step 450 loss: 0.0727\n",
            "Step 460 loss: 0.0428\n",
            "Step 470 loss: 0.0249\n",
            "Step 480 loss: 0.1095\n",
            "Step 490 loss: 0.0560\n",
            "Step 500 loss: 0.2154\n",
            "Step 510 loss: 0.0326\n",
            "Step 520 loss: 0.0948\n",
            "Step 530 loss: 0.0440\n",
            "Step 540 loss: 0.0520\n",
            "Step 550 loss: 0.0613\n",
            "Step 560 loss: 0.1523\n",
            "Step 570 loss: 0.0807\n",
            "Step 580 loss: 0.1178\n",
            "Step 590 loss: 0.0509\n",
            "Step 600 loss: 0.0813\n",
            "Step 610 loss: 0.0686\n",
            "Step 620 loss: 0.1808\n",
            "Step 630 loss: 0.0507\n",
            "Step 640 loss: 0.1384\n",
            "Step 650 loss: 0.0262\n",
            "Step 660 loss: 0.0794\n",
            "Step 670 loss: 0.0238\n",
            "Step 680 loss: 0.0962\n",
            "Step 690 loss: 0.0417\n",
            "Step 700 loss: 0.0529\n",
            "Step 710 loss: 0.0429\n",
            "Step 720 loss: 0.0503\n",
            "Step 730 loss: 0.0711\n",
            "Step 740 loss: 0.0357\n",
            "Step 750 loss: 0.1911\n",
            "Step 760 loss: 0.0686\n",
            "Step 770 loss: 0.0611\n",
            "Step 780 loss: 0.2973\n",
            "Step 790 loss: 0.1599\n",
            "Step 800 loss: 0.0410\n",
            "\n",
            "\n",
            "train_loss: 0.1081 train_rmse: 142.1004 train_mae: 87.5396\n",
            "val_loss: 0.0763 val_rmse: 103.0254 val_mae: 62.9117\n",
            "\n",
            "\n",
            "Epoch 8/10\n",
            "\n",
            "Step 0 loss: 0.2598\n",
            "Step 10 loss: 0.0799\n",
            "Step 20 loss: 0.0768\n",
            "Step 30 loss: 0.1161\n",
            "Step 40 loss: 0.1012\n",
            "Step 50 loss: 0.0761\n",
            "Step 60 loss: 0.0641\n",
            "Step 70 loss: 0.1517\n",
            "Step 80 loss: 0.1241\n",
            "Step 90 loss: 0.0767\n",
            "Step 100 loss: 0.1682\n",
            "Step 110 loss: 0.1299\n",
            "Step 120 loss: 0.0650\n",
            "Step 130 loss: 0.0594\n",
            "Step 140 loss: 0.1011\n",
            "Step 150 loss: 0.0819\n",
            "Step 160 loss: 0.1325\n",
            "Step 170 loss: 0.0528\n",
            "Step 180 loss: 0.0476\n",
            "Step 190 loss: 0.0787\n",
            "Step 200 loss: 0.1039\n",
            "Step 210 loss: 0.0564\n",
            "Step 220 loss: 0.0513\n",
            "Step 230 loss: 0.1186\n",
            "Step 240 loss: 0.2144\n",
            "Step 250 loss: 0.2551\n",
            "Step 260 loss: 0.1830\n",
            "Step 270 loss: 0.0739\n",
            "Step 280 loss: 0.1143\n",
            "Step 290 loss: 0.0644\n",
            "Step 300 loss: 0.1318\n",
            "Step 310 loss: 0.0882\n",
            "Step 320 loss: 0.0731\n",
            "Step 330 loss: 0.0483\n",
            "Step 340 loss: 0.1145\n",
            "Step 350 loss: 0.0732\n",
            "Step 360 loss: 0.0890\n",
            "Step 370 loss: 0.2060\n",
            "Step 380 loss: 0.2298\n",
            "Step 390 loss: 0.1468\n",
            "Step 400 loss: 0.1277\n",
            "Step 410 loss: 0.1126\n",
            "Step 420 loss: 0.0608\n",
            "Step 430 loss: 0.2247\n",
            "Step 440 loss: 0.1750\n",
            "Step 450 loss: 0.1671\n",
            "Step 460 loss: 0.0730\n",
            "Step 470 loss: 0.0677\n",
            "Step 480 loss: 0.0321\n",
            "Step 490 loss: 0.0626\n",
            "Step 500 loss: 0.1209\n",
            "Step 510 loss: 0.0850\n",
            "Step 520 loss: 0.0815\n",
            "Step 530 loss: 0.0997\n",
            "Step 540 loss: 0.0776\n",
            "Step 550 loss: 0.0914\n",
            "Step 560 loss: 0.0869\n",
            "Step 570 loss: 0.0658\n",
            "Step 580 loss: 0.0668\n",
            "Step 590 loss: 0.1327\n",
            "Step 600 loss: 0.1346\n",
            "Step 610 loss: 0.2091\n",
            "Step 620 loss: 0.0736\n",
            "Step 630 loss: 0.0430\n",
            "Step 640 loss: 0.0845\n",
            "Step 650 loss: 0.0690\n",
            "Step 660 loss: 0.0700\n",
            "Step 670 loss: 0.0952\n",
            "Step 680 loss: 0.0442\n",
            "Step 690 loss: 0.0525\n",
            "Step 700 loss: 0.0660\n",
            "Step 710 loss: 0.1480\n",
            "Step 720 loss: 0.0888\n",
            "Step 730 loss: 0.0869\n",
            "Step 740 loss: 0.0504\n",
            "Step 750 loss: 0.2346\n",
            "Step 760 loss: 0.0779\n",
            "Step 770 loss: 0.0820\n",
            "Step 780 loss: 0.3312\n",
            "Step 790 loss: 0.1196\n",
            "Step 800 loss: 0.6749\n",
            "\n",
            "train_loss: 0.1069 train_rmse: 140.2153 train_mae: 87.1340\n",
            "\n",
            "Step 0 loss: 0.0691\n",
            "Step 10 loss: 0.0378\n",
            "Step 20 loss: 0.1116\n",
            "Step 30 loss: 0.0714\n",
            "Step 40 loss: 0.0673\n",
            "Step 50 loss: 0.1327\n",
            "Step 60 loss: 0.2680\n",
            "Step 70 loss: 0.0996\n",
            "Step 80 loss: 0.0801\n",
            "Step 90 loss: 0.1586\n",
            "Step 100 loss: 0.1370\n",
            "Step 110 loss: 0.0387\n",
            "Step 120 loss: 0.0600\n",
            "Step 130 loss: 0.0493\n",
            "Step 140 loss: 0.0687\n",
            "Step 150 loss: 0.1071\n",
            "Step 160 loss: 0.0419\n",
            "Step 170 loss: 0.1149\n",
            "Step 180 loss: 0.0711\n",
            "Step 190 loss: 0.1544\n",
            "Step 200 loss: 0.0430\n",
            "Step 210 loss: 0.1215\n",
            "Step 220 loss: 0.0946\n",
            "Step 230 loss: 0.0830\n",
            "Step 240 loss: 0.1119\n",
            "Step 250 loss: 0.0620\n",
            "Step 260 loss: 0.0297\n",
            "Step 270 loss: 0.0551\n",
            "Step 280 loss: 0.0516\n",
            "Step 290 loss: 0.0514\n",
            "Step 300 loss: 0.0645\n",
            "Step 310 loss: 0.0439\n",
            "Step 320 loss: 0.0430\n",
            "Step 330 loss: 0.1022\n",
            "Step 340 loss: 0.0487\n",
            "Step 350 loss: 0.1851\n",
            "Step 360 loss: 0.0653\n",
            "Step 370 loss: 0.1112\n",
            "Step 380 loss: 0.0596\n",
            "Step 390 loss: 0.0416\n",
            "Step 400 loss: 0.1995\n",
            "Step 410 loss: 0.0940\n",
            "Step 420 loss: 0.1276\n",
            "Step 430 loss: 0.0415\n",
            "Step 440 loss: 0.1592\n",
            "Step 450 loss: 0.1776\n",
            "Step 460 loss: 0.0468\n",
            "Step 470 loss: 0.0392\n",
            "Step 480 loss: 0.0735\n",
            "Step 490 loss: 0.0370\n",
            "Step 500 loss: 0.0466\n",
            "Step 510 loss: 0.0758\n",
            "Step 520 loss: 0.0768\n",
            "Step 530 loss: 0.0753\n",
            "Step 540 loss: 0.0365\n",
            "Step 550 loss: 0.1704\n",
            "Step 560 loss: 0.0859\n",
            "Step 570 loss: 0.0460\n",
            "Step 580 loss: 0.1614\n",
            "Step 590 loss: 0.2557\n",
            "Step 600 loss: 0.0493\n",
            "Step 610 loss: 0.0894\n",
            "Step 620 loss: 0.0735\n",
            "Step 630 loss: 0.1364\n",
            "Step 640 loss: 0.1237\n",
            "Step 650 loss: 0.0350\n",
            "Step 660 loss: 0.0688\n",
            "Step 670 loss: 0.0474\n",
            "Step 680 loss: 0.1802\n",
            "Step 690 loss: 0.0388\n",
            "Step 700 loss: 0.2180\n",
            "Step 710 loss: 0.0425\n",
            "Step 720 loss: 0.0334\n",
            "Step 730 loss: 0.0413\n",
            "Step 740 loss: 0.0832\n",
            "Step 750 loss: 0.0420\n",
            "Step 760 loss: 0.3036\n",
            "Step 770 loss: 0.0629\n",
            "Step 780 loss: 0.0778\n",
            "Step 790 loss: 0.0383\n",
            "Step 800 loss: 0.0635\n",
            "\n",
            "\n",
            "train_loss: 0.1069 train_rmse: 140.2153 train_mae: 87.1340\n",
            "val_loss: 0.0778 val_rmse: 96.2997 val_mae: 62.9771\n",
            "\n",
            "\n",
            "Epoch 9/10\n",
            "\n",
            "Step 0 loss: 0.0938\n",
            "Step 10 loss: 0.3424\n",
            "Step 20 loss: 0.0705\n",
            "Step 30 loss: 0.0497\n",
            "Step 40 loss: 0.1400\n",
            "Step 50 loss: 0.0977\n",
            "Step 60 loss: 0.0671\n",
            "Step 70 loss: 0.1521\n",
            "Step 80 loss: 0.2080\n",
            "Step 90 loss: 0.1305\n",
            "Step 100 loss: 0.1756\n",
            "Step 110 loss: 0.3230\n",
            "Step 120 loss: 0.0719\n",
            "Step 130 loss: 0.0793\n",
            "Step 140 loss: 0.0651\n",
            "Step 150 loss: 0.1400\n",
            "Step 160 loss: 0.0536\n",
            "Step 170 loss: 0.0535\n",
            "Step 180 loss: 0.0750\n",
            "Step 190 loss: 0.0894\n",
            "Step 200 loss: 0.0740\n",
            "Step 210 loss: 0.1223\n",
            "Step 220 loss: 0.0440\n",
            "Step 230 loss: 0.0511\n",
            "Step 240 loss: 0.1229\n",
            "Step 250 loss: 0.0714\n",
            "Step 260 loss: 0.1311\n",
            "Step 270 loss: 0.0732\n",
            "Step 280 loss: 0.0794\n",
            "Step 290 loss: 0.1098\n",
            "Step 300 loss: 0.1064\n",
            "Step 310 loss: 0.0937\n",
            "Step 320 loss: 0.0505\n",
            "Step 330 loss: 0.1346\n",
            "Step 340 loss: 0.2458\n",
            "Step 350 loss: 0.0654\n",
            "Step 360 loss: 0.0656\n",
            "Step 370 loss: 0.0500\n",
            "Step 380 loss: 0.1513\n",
            "Step 390 loss: 0.0560\n",
            "Step 400 loss: 0.1046\n",
            "Step 410 loss: 0.0832\n",
            "Step 420 loss: 0.2681\n",
            "Step 430 loss: 0.0662\n",
            "Step 440 loss: 0.1351\n",
            "Step 450 loss: 0.0710\n",
            "Step 460 loss: 0.0950\n",
            "Step 470 loss: 0.0638\n",
            "Step 480 loss: 0.0885\n",
            "Step 490 loss: 0.0849\n",
            "Step 500 loss: 0.0829\n",
            "Step 510 loss: 0.0671\n",
            "Step 520 loss: 0.0778\n",
            "Step 530 loss: 0.1109\n",
            "Step 540 loss: 0.1021\n",
            "Step 550 loss: 0.1500\n",
            "Step 560 loss: 0.0701\n",
            "Step 570 loss: 0.0899\n",
            "Step 580 loss: 0.0855\n",
            "Step 590 loss: 0.0742\n",
            "Step 600 loss: 0.0574\n",
            "Step 610 loss: 0.1231\n",
            "Step 620 loss: 0.0599\n",
            "Step 630 loss: 0.0607\n",
            "Step 640 loss: 0.1057\n",
            "Step 650 loss: 0.2347\n",
            "Step 660 loss: 0.1928\n",
            "Step 670 loss: 0.1165\n",
            "Step 680 loss: 0.0961\n",
            "Step 690 loss: 0.1297\n",
            "Step 700 loss: 0.1215\n",
            "Step 710 loss: 0.2305\n",
            "Step 720 loss: 0.0806\n",
            "Step 730 loss: 0.0671\n",
            "Step 740 loss: 0.1189\n",
            "Step 750 loss: 0.1317\n",
            "Step 760 loss: 0.2028\n",
            "Step 770 loss: 0.0790\n",
            "Step 780 loss: 0.0526\n",
            "Step 790 loss: 0.0850\n",
            "Step 800 loss: 0.3732\n",
            "\n",
            "train_loss: 0.1023 train_rmse: 138.8674 train_mae: 85.5004\n",
            "\n",
            "Step 0 loss: 0.0468\n",
            "Step 10 loss: 0.0577\n",
            "Step 20 loss: 0.0955\n",
            "Step 30 loss: 0.3955\n",
            "Step 40 loss: 0.0841\n",
            "Step 50 loss: 0.0558\n",
            "Step 60 loss: 0.0860\n",
            "Step 70 loss: 0.1227\n",
            "Step 80 loss: 0.0915\n",
            "Step 90 loss: 0.4040\n",
            "Step 100 loss: 0.0434\n",
            "Step 110 loss: 0.0757\n",
            "Step 120 loss: 0.0603\n",
            "Step 130 loss: 0.1498\n",
            "Step 140 loss: 0.0598\n",
            "Step 150 loss: 0.1425\n",
            "Step 160 loss: 0.1669\n",
            "Step 170 loss: 0.0524\n",
            "Step 180 loss: 0.0301\n",
            "Step 190 loss: 0.0391\n",
            "Step 200 loss: 0.0906\n",
            "Step 210 loss: 0.1508\n",
            "Step 220 loss: 0.0358\n",
            "Step 230 loss: 0.0879\n",
            "Step 240 loss: 0.0694\n",
            "Step 250 loss: 0.0517\n",
            "Step 260 loss: 0.1411\n",
            "Step 270 loss: 0.0388\n",
            "Step 280 loss: 0.0376\n",
            "Step 290 loss: 0.0206\n",
            "Step 300 loss: 0.0668\n",
            "Step 310 loss: 0.1109\n",
            "Step 320 loss: 0.0553\n",
            "Step 330 loss: 0.0395\n",
            "Step 340 loss: 0.3328\n",
            "Step 350 loss: 0.0506\n",
            "Step 360 loss: 0.1503\n",
            "Step 370 loss: 0.0304\n",
            "Step 380 loss: 0.0523\n",
            "Step 390 loss: 0.0538\n",
            "Step 400 loss: 0.0736\n",
            "Step 410 loss: 0.0995\n",
            "Step 420 loss: 0.0496\n",
            "Step 430 loss: 0.1091\n",
            "Step 440 loss: 0.0447\n",
            "Step 450 loss: 0.0260\n",
            "Step 460 loss: 0.0266\n",
            "Step 470 loss: 0.0739\n",
            "Step 480 loss: 0.0997\n",
            "Step 490 loss: 0.1138\n",
            "Step 500 loss: 0.0977\n",
            "Step 510 loss: 0.0635\n",
            "Step 520 loss: 0.0777\n",
            "Step 530 loss: 0.0624\n",
            "Step 540 loss: 0.0352\n",
            "Step 550 loss: 0.0382\n",
            "Step 560 loss: 0.1734\n",
            "Step 570 loss: 0.0292\n",
            "Step 580 loss: 0.1786\n",
            "Step 590 loss: 0.0692\n",
            "Step 600 loss: 0.0615\n",
            "Step 610 loss: 0.0437\n",
            "Step 620 loss: 0.0378\n",
            "Step 630 loss: 0.0811\n",
            "Step 640 loss: 0.0542\n",
            "Step 650 loss: 0.0613\n",
            "Step 660 loss: 0.0517\n",
            "Step 670 loss: 0.0846\n",
            "Step 680 loss: 0.0520\n",
            "Step 690 loss: 0.1612\n",
            "Step 700 loss: 0.1170\n",
            "Step 710 loss: 0.0440\n",
            "Step 720 loss: 0.1854\n",
            "Step 730 loss: 0.0459\n",
            "Step 740 loss: 0.3419\n",
            "Step 750 loss: 0.0501\n",
            "Step 760 loss: 0.1072\n",
            "Step 770 loss: 0.0319\n",
            "Step 780 loss: 0.0823\n",
            "Step 790 loss: 0.0797\n",
            "Step 800 loss: 0.0577\n",
            "\n",
            "\n",
            "train_loss: 0.1023 train_rmse: 138.8674 train_mae: 85.5004\n",
            "val_loss: 0.0768 val_rmse: 96.2968 val_mae: 65.7494\n",
            "\n",
            "\n",
            "Epoch 10/10\n",
            "\n",
            "Step 0 loss: 0.0601\n",
            "Step 10 loss: 0.0522\n",
            "Step 20 loss: 0.0607\n",
            "Step 30 loss: 0.0854\n",
            "Step 40 loss: 0.1246\n",
            "Step 50 loss: 0.1012\n",
            "Step 60 loss: 0.0720\n",
            "Step 70 loss: 0.0656\n",
            "Step 80 loss: 0.0574\n",
            "Step 90 loss: 0.2251\n",
            "Step 100 loss: 0.0603\n",
            "Step 110 loss: 0.0746\n",
            "Step 120 loss: 0.0605\n",
            "Step 130 loss: 0.0628\n",
            "Step 140 loss: 0.1011\n",
            "Step 150 loss: 0.0593\n",
            "Step 160 loss: 0.1457\n",
            "Step 170 loss: 0.0786\n",
            "Step 180 loss: 0.0480\n",
            "Step 190 loss: 0.1200\n",
            "Step 200 loss: 0.0536\n",
            "Step 210 loss: 0.0839\n",
            "Step 220 loss: 0.0748\n",
            "Step 230 loss: 0.0609\n",
            "Step 240 loss: 0.1021\n",
            "Step 250 loss: 0.1269\n",
            "Step 260 loss: 0.0648\n",
            "Step 270 loss: 0.0642\n",
            "Step 280 loss: 0.1498\n",
            "Step 290 loss: 0.0988\n",
            "Step 300 loss: 0.0861\n",
            "Step 310 loss: 0.0442\n",
            "Step 320 loss: 0.1351\n",
            "Step 330 loss: 0.0513\n",
            "Step 340 loss: 0.0822\n",
            "Step 350 loss: 0.0646\n",
            "Step 360 loss: 0.1037\n",
            "Step 370 loss: 0.0847\n",
            "Step 380 loss: 0.2267\n",
            "Step 390 loss: 0.0657\n",
            "Step 400 loss: 0.0742\n",
            "Step 410 loss: 0.0756\n",
            "Step 420 loss: 0.2018\n",
            "Step 430 loss: 0.1256\n",
            "Step 440 loss: 0.0727\n",
            "Step 450 loss: 0.0849\n",
            "Step 460 loss: 0.0822\n",
            "Step 470 loss: 0.0730\n",
            "Step 480 loss: 0.0577\n",
            "Step 490 loss: 0.0562\n",
            "Step 500 loss: 0.0832\n",
            "Step 510 loss: 0.0714\n",
            "Step 520 loss: 0.0742\n",
            "Step 530 loss: 0.0689\n",
            "Step 540 loss: 0.0718\n",
            "Step 550 loss: 0.0813\n",
            "Step 560 loss: 0.1043\n",
            "Step 570 loss: 0.0756\n",
            "Step 580 loss: 0.0944\n",
            "Step 590 loss: 0.0742\n",
            "Step 600 loss: 0.0899\n",
            "Step 610 loss: 0.0874\n",
            "Step 620 loss: 0.0854\n",
            "Step 630 loss: 0.0758\n",
            "Step 640 loss: 0.0851\n",
            "Step 650 loss: 0.0803\n",
            "Step 660 loss: 0.1366\n",
            "Step 670 loss: 0.0757\n",
            "Step 680 loss: 0.1165\n",
            "Step 690 loss: 0.0950\n",
            "Step 700 loss: 0.1333\n",
            "Step 710 loss: 0.0583\n",
            "Step 720 loss: 0.0580\n",
            "Step 730 loss: 0.1843\n",
            "Step 740 loss: 0.0556\n",
            "Step 750 loss: 0.0577\n",
            "Step 760 loss: 0.0900\n",
            "Step 770 loss: 0.2020\n",
            "Step 780 loss: 0.1416\n",
            "Step 790 loss: 0.0725\n",
            "Step 800 loss: 0.1728\n",
            "\n",
            "train_loss: 0.0970 train_rmse: 133.4701 train_mae: 82.8143\n",
            "\n",
            "Step 0 loss: 0.0350\n",
            "Step 10 loss: 0.0612\n",
            "Step 20 loss: 0.0995\n",
            "Step 30 loss: 0.0905\n",
            "Step 40 loss: 0.0764\n",
            "Step 50 loss: 0.2495\n",
            "Step 60 loss: 0.0457\n",
            "Step 70 loss: 0.0317\n",
            "Step 80 loss: 0.0461\n",
            "Step 90 loss: 0.1614\n",
            "Step 100 loss: 0.0556\n",
            "Step 110 loss: 0.1267\n",
            "Step 120 loss: 0.0285\n",
            "Step 130 loss: 0.1240\n",
            "Step 140 loss: 0.0331\n",
            "Step 150 loss: 0.0683\n",
            "Step 160 loss: 0.0525\n",
            "Step 170 loss: 0.0518\n",
            "Step 180 loss: 0.0545\n",
            "Step 190 loss: 0.2548\n",
            "Step 200 loss: 0.0505\n",
            "Step 210 loss: 0.0634\n",
            "Step 220 loss: 0.0524\n",
            "Step 230 loss: 0.0778\n",
            "Step 240 loss: 0.0235\n",
            "Step 250 loss: 0.0456\n",
            "Step 260 loss: 0.0837\n",
            "Step 270 loss: 0.0961\n",
            "Step 280 loss: 0.0691\n",
            "Step 290 loss: 0.0181\n",
            "Step 300 loss: 0.1271\n",
            "Step 310 loss: 0.0662\n",
            "Step 320 loss: 0.1410\n",
            "Step 330 loss: 0.1180\n",
            "Step 340 loss: 0.1498\n",
            "Step 350 loss: 0.0909\n",
            "Step 360 loss: 0.0534\n",
            "Step 370 loss: 0.0366\n",
            "Step 380 loss: 0.0591\n",
            "Step 390 loss: 0.0675\n",
            "Step 400 loss: 0.0516\n",
            "Step 410 loss: 0.0587\n",
            "Step 420 loss: 0.0412\n",
            "Step 430 loss: 0.1243\n",
            "Step 440 loss: 0.0509\n",
            "Step 450 loss: 0.0632\n",
            "Step 460 loss: 0.0399\n",
            "Step 470 loss: 0.0608\n",
            "Step 480 loss: 0.0271\n",
            "Step 490 loss: 0.0506\n",
            "Step 500 loss: 0.0489\n",
            "Step 510 loss: 0.1114\n",
            "Step 520 loss: 0.0487\n",
            "Step 530 loss: 0.0613\n",
            "Step 540 loss: 0.0350\n",
            "Step 550 loss: 0.3232\n",
            "Step 560 loss: 0.0517\n",
            "Step 570 loss: 0.0419\n",
            "Step 580 loss: 0.0411\n",
            "Step 590 loss: 0.0469\n",
            "Step 600 loss: 0.0657\n",
            "Step 610 loss: 0.0768\n",
            "Step 620 loss: 0.0467\n",
            "Step 630 loss: 0.0389\n",
            "Step 640 loss: 0.0263\n",
            "Step 650 loss: 0.0481\n",
            "Step 660 loss: 0.0877\n",
            "Step 670 loss: 0.0546\n",
            "Step 680 loss: 0.0397\n",
            "Step 690 loss: 0.1560\n",
            "Step 700 loss: 0.0399\n",
            "Step 710 loss: 0.3562\n",
            "Step 720 loss: 0.0412\n",
            "Step 730 loss: 0.0606\n",
            "Step 740 loss: 0.0344\n",
            "Step 750 loss: 0.0575\n",
            "Step 760 loss: 0.0376\n",
            "Step 770 loss: 0.1171\n",
            "Step 780 loss: 0.0471\n",
            "Step 790 loss: 0.0504\n",
            "Step 800 loss: 0.0575\n",
            "\n",
            "\n",
            "train_loss: 0.0970 train_rmse: 133.4701 train_mae: 82.8143\n",
            "val_loss: 0.0664 val_rmse: 90.3003 val_mae: 57.7842\n",
            "\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q9b4xoxYIQf8"
      },
      "source": [
        "trainer.restore_best()"
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UCgA-pwFDqeD"
      },
      "source": [
        "# Visualize Model's Predictions"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "259-UVoPE_om"
      },
      "source": [
        "## rmse of model that predicts mean of cycle_life"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v2ZMtlzH8070"
      },
      "source": [
        "lives = []\n",
        "for i in range(len(train_ds)):\n",
        "  lives.append(train_ds[i][1].item())\n",
        "lives = torch.tensor(lives)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0A5aqTkX9KWA",
        "outputId": "687b6f8f-b474-44b7-9f8d-a806f35833d1"
      },
      "source": [
        "rmse(lives, torch.full((len(lives), ), torch.mean(lives)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor(385.8495)"
            ]
          },
          "metadata": {},
          "execution_count": 70
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pboLsWu2FSXk"
      },
      "source": [
        "## visualize our model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yCnJHq7xAM34"
      },
      "source": [
        "def predict(model, dataloader):\n",
        "  ypred = []\n",
        "  ytrue = []\n",
        "  with torch.no_grad():\n",
        "    for x, y in dataloader:\n",
        "      y_pred_batch = model(x)\n",
        "      ytrue.append(y)\n",
        "      ypred.append(y_pred_batch)\n",
        "  ypred =torch.cat(ypred, dim=0).flatten()\n",
        "  ytrue =torch.cat(ytrue, dim=0).flatten()\n",
        "  return ypred, ytrue\n",
        "\n",
        "def report(ypred, ytrue):\n",
        "  ypred_exp = torch.exp(ypred).detach().cpu().numpy()\n",
        "  ytrue_exp = torch.exp(ytrue).detach().cpu().numpy()\n",
        "\n",
        "  observations = len(ypred_exp)\n",
        "  mae_val = metrics.mean_absolute_error(ytrue_exp, ypred_exp)\n",
        "  rmse_val = metrics.mean_squared_error(ytrue_exp, ypred_exp, squared=False)\n",
        "  r2_val = metrics.r2_score(ytrue_exp, ypred_exp)\n",
        "  max_error_val = metrics.max_error(ytrue_exp, ypred_exp)\n",
        "\n",
        "  return pd.Series({\n",
        "      'mean_absolute_error' : mae_val, \n",
        "      'root_mean_square_error' : rmse_val, \n",
        "      'r_squared' : r2_val,\n",
        "      'max_error': max_error_val,\n",
        "      'observations': observations,\n",
        "  })\n",
        "\n",
        "def plot_residual_histogram(ypred, ytrue):\n",
        "  ypred_exp = torch.exp(ypred).detach().cpu().numpy()\n",
        "  ytrue_exp = torch.exp(ytrue).detach().cpu().numpy()\n",
        "  residual = np.abs(ypred_exp - ytrue_exp)\n",
        "  return sns.histplot(x=residual)\n",
        "\n",
        "def plot_predictions(ypred, ytrue):\n",
        "  plt.plot(ytrue.detach().cpu().numpy(), label='actual')\n",
        "  plt.plot(ypred.detach().cpu().numpy(), label='pred')\n",
        "  plt.legend()"
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QwrWkIVw2Q3l"
      },
      "source": [
        "train set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eYnCwxdkC3ni"
      },
      "source": [
        "ypred, ytrue = predict(model, train_dl)"
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hwQHLlYQ2AXX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "outputId": "643944ce-006b-4f89-d9f0-7fcacf55c97c"
      },
      "source": [
        "plot_residual_histogram(ypred, ytrue).set_title('Train Residuals Histogram');"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HRT2ZUOZ1wu9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "outputId": "29972030-d768-4e33-83be-7772fa051b72"
      },
      "source": [
        "plot_predictions(ypred[:100], ytrue[:100])\n",
        "plt.title(\"Few Predictions On Training Data\");"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1p54a9KdC8l1",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8de19491-9c64-46e2-a6e8-3db0e0c165c5"
      },
      "source": [
        "print(\"Train Set Report\")\n",
        "report(ypred, ytrue)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Train Set Report\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "mean_absolute_error          57.786537\n",
              "root_mean_square_error       90.318970\n",
              "r_squared                     0.936635\n",
              "max_error                   882.436218\n",
              "observations              25830.000000\n",
              "dtype: float64"
            ]
          },
          "metadata": {},
          "execution_count": 31
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yklsiuwG2W4V"
      },
      "source": [
        "validation set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_XRVUI67JXGv"
      },
      "source": [
        "ypred, ytrue = predict(model, val_dl)"
      ],
      "execution_count": 32,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IwLHlNP22bZm",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "outputId": "ebb03b4a-74e1-44f5-9aab-b60534bcf6a6"
      },
      "source": [
        "plot_residual_histogram(ypred, ytrue).set_title('Validation Residuals Histogram');"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "We-BgAM02c4P",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "outputId": "ee3a45b3-0727-452d-b789-5469afdba2d0"
      },
      "source": [
        "plot_predictions(ypred[:100], ytrue[:100])\n",
        "plt.title(\"Few Predictions On Validation Data\");"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Is5cxN2R2fHk",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "607e045f-ae93-46d9-d249-7bf5456abdd2"
      },
      "source": [
        "print(\"Validation Set Report\")\n",
        "report(ypred, ytrue)"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Validation Set Report\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "mean_absolute_error          78.637405\n",
              "root_mean_square_error      130.918839\n",
              "r_squared                     0.897482\n",
              "max_error                  1049.312744\n",
              "observations              29454.000000\n",
              "dtype: float64"
            ]
          },
          "metadata": {},
          "execution_count": 35
        }
      ]
    }
  ]
}