A3.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import unicode_literals, print_function, division\n",
    "from io import open\n",
    "import unicodedata\n",
    "import string\n",
    "import re\n",
    "import random\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "from torch import optim\n",
    "import torch.nn.functional as F\n",
    "import wandb \n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Failed to detect the name of this notebook, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable to enable code saving.\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mcs20b004\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n"
     ]
    }
   ],
   "source": [
    "SOS_token = 0\n",
    "EOS_token = 1\n",
    "MAX_LENGTH = 40\n",
    "WANDB_NOTEBOOK_NAME = 'Assignment3'\n",
    "WANDB_PROJECT_NAME = 'CS6910_A3'\n",
    "WANDB_ENTITY = 'cs20b004'\n",
    "model_number = 0\n",
    "wandb.login()\n",
    "\n",
    "class Lang:\n",
    "    def __init__(self, name):\n",
    "        self.name = name\n",
    "        # self.word2index = {}\n",
    "        # self.word2count = {}\n",
    "        # self.index2word = {0: \"SOS\", 1: \"EOS\"}\n",
    "        # self.n_words = 2  # Count SOS and EOS\n",
    "        self.letter_index = {}\n",
    "        self.letter_count = {}\n",
    "        self.index_letter = {0: \"SOS\", 1: \"EOS\"}\n",
    "        self.n_letters = 2  # Count SOS and EOS\n",
    "\n",
    "\n",
    "    def add_word(self, word):\n",
    "        # for word in sentence.split(' '):\n",
    "        #     self.addWord(word)\n",
    "        for letter in word:\n",
    "            self.add_letter(letter)\n",
    "\n",
    "    def add_letter(self, letter):\n",
    "        # if word not in self.word2index:\n",
    "        #     self.word2index[word] = self.n_words\n",
    "        #     self.word2count[word] = 1\n",
    "        #     self.index2word[self.n_words] = word\n",
    "        #     self.n_words += 1\n",
    "        # else:\n",
    "        #     self.word2count[word] += 1\n",
    "        if letter not in self.letter_index:\n",
    "            self.letter_index[letter] = self.n_letters\n",
    "            self.letter_count[letter] = 1\n",
    "            self.index_letter[self.n_letters] = letter\n",
    "            self.n_letters += 1\n",
    "        else:\n",
    "            self.letter_count[letter] += 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def read_words(output_lang, type = 'train', reverse = False):\n",
    "    lines = open('aksharantar_sampled/%s/%s_%s.csv' % (output_lang,output_lang,type), encoding='utf-8').read().strip().split('\\n')\n",
    "    pairs = [[s for s in l.split(',')] for l in lines]\n",
    "    if reverse:\n",
    "        pairs = [list(reversed(p)) for p in pairs]\n",
    "    else:\n",
    "        pairs = [list(p) for p in pairs]\n",
    "    return pairs\n",
    "\n",
    "def init_lang(lang1, lang2, type = 'train', reverse = False):\n",
    "    pairs = read_words(lang2, type, reverse)\n",
    "    input_lang = Lang(lang1)\n",
    "    output_lang = Lang(lang2)\n",
    "    char = 'a'\n",
    "    for i in range(26):\n",
    "        input_lang.add_letter(char)\n",
    "        char = chr(ord(char) + 1)\n",
    "    char = '\\u0900'\n",
    "    for i in range(128):\n",
    "        output_lang.add_letter(char)\n",
    "        char = chr(ord(char) + 1)\n",
    "    # for pair in pairs:\n",
    "    #     input_lang.add_word(pair[0])\n",
    "    #     output_lang.add_word(pair[1])\n",
    "    \n",
    "    return input_lang, output_lang, pairs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MAX_LENGTH:  40\n"
     ]
    }
   ],
   "source": [
    "pairs = read_words('mar',type = 'train')\n",
    "for p in pairs:\n",
    "    MAX_LENGTH = max(MAX_LENGTH, len(p[1]))\n",
    "print(\"MAX_LENGTH: \", MAX_LENGTH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# def indexesFromSentence(lang, sentence):\n",
    "#     return [lang.word2index[word] for word in sentence.split(' ')]\n",
    "def word_to_index(lang, word):\n",
    "    return [lang.letter_index[letter] for letter in word]\n",
    "\n",
    "\n",
    "# def tensorFromSentence(lang, sentence):\n",
    "#     indexes = indexesFromSentence(lang, sentence)\n",
    "#     indexes.append(EOS_token)\n",
    "#     return torch.tensor(indexes, dtype=torch.long, device=device).view(-1, 1)\n",
    "\n",
    "def word_to_tensor(lang, word):\n",
    "    indexes = word_to_index(lang, word)\n",
    "    indexes.append(EOS_token)\n",
    "    return torch.tensor(indexes, dtype=torch.long, device=device).view(-1, 1)\n",
    "\n",
    "\n",
    "# def tensorsFromPair(pair):\n",
    "#     input_tensor = tensorFromSentence(input_lang, pair[0])\n",
    "#     target_tensor = tensorFromSentence(output_lang, pair[1])\n",
    "#     return (input_tensor, target_tensor)\n",
    "\n",
    "def pair_to_tensor(input_lang,output_lang, pair):\n",
    "    input_tensor = word_to_tensor(input_lang, pair[0])\n",
    "    target_tensor = word_to_tensor(output_lang, pair[1])\n",
    "    return (input_tensor, target_tensor)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class EncoderRNN(nn.Module):\n",
    "    def __init__(self, input_size, embedding_size, hidden_size, num_layers, type = 'gru', nonlinearity = 'tanh', dropout_p = 0.1, bidirectional = False):\n",
    "        super(EncoderRNN, self).__init__()\n",
    "        self.hidden_size = hidden_size\n",
    "        self.embedding_size = embedding_size\n",
    "        self.num_layers = num_layers\n",
    "        self.embedding = nn.Embedding(input_size, embedding_size)\n",
    "        self.dropout_p = dropout_p\n",
    "        self.bidirectional = bidirectional\n",
    "        self.dropout = nn.Dropout(self.dropout_p)\n",
    "        self.type = type\n",
    "        if self.type == 'gru':\n",
    "            self.gru = nn.GRU(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'lstm':\n",
    "            self.lstm = nn.LSTM(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'rnn':\n",
    "            self.rnn = nn.RNN(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "\n",
    "    def forward(self, input, hidden, cell):\n",
    "        embedded = self.embedding(input).view(1, 1, -1)\n",
    "        output = embedded\n",
    "        if self.type == 'gru':\n",
    "            output, hidden = self.gru(output, hidden)\n",
    "        elif self.type == 'lstm':\n",
    "            output, (hidden, cell) = self.lstm(output, (hidden, cell))\n",
    "        elif self.type == 'rnn':\n",
    "            output, hidden = self.rnn(output, hidden)\n",
    "        return output, hidden, cell\n",
    "\n",
    "    def initHidden(self):\n",
    "        return torch.zeros(self.num_layers , 1, self.hidden_size, device=device)\n",
    "    \n",
    "    def init_cell(self):\n",
    "        return self.initHidden()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DecoderRNN(nn.Module):\n",
    "    def __init__(self, hidden_size, embedding_size, output_size, num_layers, nonlinearity = 'tanh', dropout_p = 0.1, type = 'gru', bidirectional = False):\n",
    "        super(DecoderRNN, self).__init__()\n",
    "        self.hidden_size = hidden_size\n",
    "        self.embedding_size = embedding_size\n",
    "        self.dropout_p = dropout_p\n",
    "        self.embedding = nn.Embedding(output_size, embedding_size)\n",
    "        # self.dropout = nn.Dropout(self.dropout_p)\n",
    "        self.bidirectional = bidirectional\n",
    "        self.num_layers = num_layers\n",
    "        self.dropout = nn.Dropout(self.dropout_p)\n",
    "        self.type = type\n",
    "        if self.type == 'gru':\n",
    "            self.gru = nn.GRU(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'lstm':\n",
    "            self.lstm = nn.LSTM(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'rnn':\n",
    "            self.rnn = nn.RNN(embedding_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "            \n",
    "        self.out = nn.Linear(hidden_size, output_size)\n",
    "        self.softmax = nn.LogSoftmax(dim=1)\n",
    "\n",
    "    def forward(self, input, hidden, cell):\n",
    "        output = self.embedding(input).view(1, 1, -1)\n",
    "        output = self.dropout(output)\n",
    "        output = F.relu(output)\n",
    "        if self.type == 'gru':\n",
    "            output, hidden = self.gru(output, hidden)\n",
    "        elif self.type == 'lstm':\n",
    "            output, (hidden, cell) = self.lstm(output, (hidden, cell))\n",
    "        elif self.type == 'rnn':\n",
    "            output, hidden = self.rnn(output, hidden)\n",
    "        \n",
    "        output = self.softmax(self.out(output[0]))\n",
    "        return output, hidden, cell\n",
    "\n",
    "    def initHidden(self):\n",
    "        return torch.zeros(self.num_layers, 1, self.hidden_size, device=device)\n",
    "    \n",
    "        \n",
    "    def init_cell(self):\n",
    "        return self.initHidden()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "class AttnDecoderRNN(nn.Module):\n",
    "    def __init__(self, hidden_size, embedding_size, output_size, num_layers, nonlinearity = 'tanh', dropout_p=0.1 , max_length=MAX_LENGTH, type = 'gru', bidirectional = False):\n",
    "        super(AttnDecoderRNN, self).__init__()\n",
    "        self.hidden_size = hidden_size\n",
    "        self.embedding_size = embedding_size\n",
    "        self.output_size = output_size\n",
    "        self.dropout_p = dropout_p\n",
    "        self.max_length = max_length\n",
    "        self.num_layers = num_layers\n",
    "        self.embedding = nn.Embedding(self.output_size, self.embedding_size)\n",
    "        self.attn = nn.Linear(self.hidden_size + self.embedding_size, self.max_length)\n",
    "        self.attn_combine = nn.Linear(self.hidden_size + self.embedding_size, self.hidden_size)\n",
    "        self.dropout = nn.Dropout(self.dropout_p)\n",
    "        self.bidirectional = bidirectional\n",
    "        self.type = type\n",
    "        \n",
    "        if self.type == 'gru':\n",
    "            self.gru = nn.GRU(hidden_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'lstm':\n",
    "            self.lstm = nn.LSTM(hidden_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "        elif self.type == 'rnn':\n",
    "            self.rnn = nn.RNN(hidden_size, hidden_size, num_layers = num_layers, bidirectional = self.bidirectional)\n",
    "\n",
    "        self.out = nn.Linear(self.hidden_size, self.output_size)\n",
    "\n",
    "    def forward(self, input, hidden, cell, encoder_outputs):\n",
    "        embedded = self.embedding(input).view(1, 1, -1)\n",
    "        embedded = self.dropout(embedded)\n",
    "\n",
    "        attn_weights = F.softmax(\n",
    "            self.attn(torch.cat((embedded[0], hidden[0]), 1)), dim=1)\n",
    "        attn_applied = torch.bmm(attn_weights.unsqueeze(0),\n",
    "                                 encoder_outputs.unsqueeze(0))\n",
    "\n",
    "        output = torch.cat((embedded[0], attn_applied[0]), 1)\n",
    "        output = self.attn_combine(output).unsqueeze(0)\n",
    "\n",
    "        output = F.relu(output)\n",
    "        if self.type == 'gru':\n",
    "            output, hidden = self.gru(output, hidden)\n",
    "        elif self.type == 'lstm':\n",
    "            output, (hidden, cell) = self.lstm(output, (hidden, cell))\n",
    "        elif self.type == 'rnn':\n",
    "            output, hidden = self.rnn(output, hidden)\n",
    "\n",
    "        output = F.log_softmax(self.out(output[0]), dim=1)\n",
    "        return output, hidden, cell, attn_weights\n",
    "\n",
    "    def initHidden(self):\n",
    "        return torch.zeros(self.num_layers, 1, self.hidden_size, device=device)\n",
    "        \n",
    "    def init_cell(self):\n",
    "        return self.initHidden()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Transliterator():\n",
    "    def __init__(self, encoder_hp, decoder_hp, attn = True):\n",
    "        self.encoder = EncoderRNN(\n",
    "                        encoder_hp['input_size'],\n",
    "                        encoder_hp['embedding_size'],\n",
    "                        encoder_hp['hidden_size'], \n",
    "                        encoder_hp['num_layers'], \n",
    "                        type = encoder_hp['type'],\n",
    "                        dropout_p= encoder_hp['dropout_p'],\n",
    "                        bidirectional = encoder_hp['bidirectional']\n",
    "                        ).to(device)\n",
    "        self.attn = attn\n",
    "        if attn:\n",
    "            self.decoder = AttnDecoderRNN(\n",
    "                            decoder_hp['hidden_size'], \n",
    "                            decoder_hp['embedding_size'],\n",
    "                            decoder_hp['output_size'], \n",
    "                            decoder_hp['num_layers'], \n",
    "                            type = decoder_hp['type'],\n",
    "                            dropout_p = decoder_hp['dropout_p'],\n",
    "                            bidirectional = decoder_hp['bidirectional']\n",
    "                            ).to(device)\n",
    "        else:\n",
    "            self.decoder = DecoderRNN(\n",
    "                            decoder_hp['hidden_size'], \n",
    "                            decoder_hp['embedding_size'],\n",
    "                            decoder_hp['output_size'], \n",
    "                            decoder_hp['num_layers'],  \n",
    "                            type = decoder_hp['type'],\n",
    "                            dropout_p = decoder_hp['dropout_p'],\n",
    "                            bidirectional = decoder_hp['bidirectional']\n",
    "                            ).to(device)\n",
    "\n",
    "    def train(self, input_tensor, target_tensor, max_length=MAX_LENGTH):\n",
    "        encoder_hidden = self.encoder.initHidden()\n",
    "        encoder_cell = self.encoder.init_cell()\n",
    "\n",
    "        self.encoder_optimizer.zero_grad()\n",
    "        self.decoder_optimizer.zero_grad()\n",
    "\n",
    "        input_length = input_tensor.size(0)\n",
    "        target_length = target_tensor.size(0)\n",
    "\n",
    "        encoder_outputs = torch.zeros(max_length, self.encoder.hidden_size, device=device)\n",
    "\n",
    "        loss = 0\n",
    "\n",
    "        for ei in range(input_length):\n",
    "            encoder_output, encoder_hidden, encoder_cell = self.encoder(input_tensor[ei], encoder_hidden, encoder_cell)\n",
    "            encoder_outputs[ei] = encoder_output[0, 0]\n",
    "\n",
    "        decoder_input = torch.tensor([[SOS_token]], device=device)\n",
    "        decoder_hidden = encoder_hidden[encoder_hidden.shape[0] - 1].unsqueeze(0)\n",
    "        decoder_cell = encoder_cell[encoder_cell.shape[0] - 1].unsqueeze(0)\n",
    "\n",
    "        decoded_word = ''\n",
    "        for i in range(self.decoder.num_layers - 1):\n",
    "            decoder_hidden = torch.cat((decoder_hidden, encoder_hidden[encoder_hidden.shape[0] - 1].unsqueeze(0)), 0)\n",
    "            decoder_cell = torch.cat((decoder_cell, encoder_cell[encoder_cell.shape[0] - 1].unsqueeze(0)), 0)\n",
    "\n",
    "        use_teacher_forcing = True if random.random() < self.teacher_forcing_ratio else False\n",
    "        # print(encoder_outputs.shape)\n",
    "        if use_teacher_forcing:\n",
    "            for di in range(target_length):\n",
    "                if self.attn:\n",
    "                    decoder_output, decoder_hidden, decoder_cell, decoder_attention = self.decoder(decoder_input, decoder_hidden, decoder_cell, encoder_outputs)\n",
    "                else:\n",
    "                    decoder_output, decoder_hidden, decoder_cell = self.decoder(decoder_input, decoder_hidden, decoder_cell)\n",
    "                topv, topi = decoder_output.topk(1)\n",
    "                decoder_input = topi.squeeze().detach()  \n",
    "                loss += self.criterion(decoder_output, target_tensor[di])\n",
    "                decoded_word += self.output_lang.index_letter[topi.item()]\n",
    "                decoder_input = target_tensor[di]\n",
    "        else:\n",
    "            for di in range(target_length):\n",
    "                if self.attn:\n",
    "                    decoder_output, decoder_hidden, decoder_cell, decoder_attention = self.decoder(decoder_input, decoder_hidden, decoder_cell, encoder_outputs)\n",
    "                else:\n",
    "                    decoder_output, decoder_hidden, decoder_cell = self.decoder(decoder_input, decoder_cell, decoder_hidden)\n",
    "                topv, topi = decoder_output.topk(1)\n",
    "                decoder_input = topi.squeeze().detach()  \n",
    "                loss += self.criterion(decoder_output, target_tensor[di])\n",
    "                if decoder_input.item() == EOS_token:\n",
    "                    break\n",
    "                decoded_word += self.output_lang.index_letter[topi.item()]\n",
    "                \n",
    "        \n",
    "        loss.backward()\n",
    "\n",
    "        self.encoder_optimizer.step()\n",
    "        self.decoder_optimizer.step()\n",
    "        \n",
    "        return loss.item() / target_length, decoded_word \n",
    "    \n",
    "    def fit( self, train_io_pair, val_io_pair, input_lang, output_lang, n_epochs,optimizer = 'adam', criterion = nn.NLLLoss(), print_every=1000, learning_rate=0.0001, teacher_forcing_ratio = 0.5, use_wanb = False, model_name = 'model'):\n",
    "        self.input_lang = input_lang\n",
    "        self.output_lang = output_lang\n",
    "\n",
    "        if(optimizer == 'adam'):\n",
    "            self.encoder_optimizer = optim.Adam(self.encoder.parameters(), lr=learning_rate)\n",
    "            self.decoder_optimizer = optim.Adam(self.decoder.parameters(), lr=learning_rate)\n",
    "        elif(optimizer == 'sgd'):\n",
    "            self.encoder_optimizer = optim.SGD(self.encoder.parameters(), lr=learning_rate)\n",
    "            self.decoder_optimizer = optim.SGD(self.decoder.parameters(), lr=learning_rate)\n",
    "            \n",
    "        self.criterion = criterion\n",
    "        self.learning_rate = learning_rate\n",
    "        self.teacher_forcing_ratio = teacher_forcing_ratio\n",
    "        self.n_epochs = n_epochs\n",
    "        self.print_every = print_every\n",
    "        \n",
    "        self.all_losses = []\n",
    "        self.total_loss = 0\n",
    "\n",
    "        # for i in range(len(train_io_pair)):\n",
    "        #     train_io_pair[i] = pair_to_tensor(train_io_pair[i])\n",
    "        training_pairs = [pair_to_tensor(self.input_lang, self.output_lang, train_io_pair[i]) for i in range(len(train_io_pair))]\n",
    "        train_accs = []\n",
    "        val_accs = []\n",
    "        for epc in range(1,n_epochs + 1):\n",
    "            correct = 0\n",
    "            for inp in range(len(train_io_pair)):\n",
    "                input_tensor = training_pairs[inp][0]\n",
    "                target_tensor = training_pairs[inp][1]\n",
    "                loss, decoder_output = self.train(input_tensor, target_tensor)\n",
    "                self.total_loss += loss\n",
    "                self.all_losses.append(loss)\n",
    "                if(decoder_output == train_io_pair[inp][1]):\n",
    "                    correct += 1\n",
    "            train_acc,train_ouput = self.eval(train_io_pair)\n",
    "            train_accs.append(train_acc)\n",
    "            val_acc,val_output = self.eval(val_io_pair)\n",
    "            val_accs.append(val_acc)\n",
    "            print(\"Epoch: \", epc, \" Loss: \", self.total_loss / len(train_io_pair),\n",
    "                  \"Training Accuracy: \", train_acc,\n",
    "                  \"Validation Accuracy: \", val_acc)\n",
    "            if(use_wanb):\n",
    "                wandb.log({\n",
    "                    \"epoch\" : epc,\n",
    "                    \"train_loss\" : self.total_loss / len(train_io_pair),\n",
    "                    \"Training Accuracy\": train_acc, \n",
    "                    \"Validation Accuracy\": val_acc\n",
    "                    })\n",
    "            self.total_loss = 0\n",
    "            if epc >5 and val_acc < 0.3:\n",
    "                break\n",
    "            if len(val_accs) > 3 and val_accs[-1] < val_accs[-3]:\n",
    "                break\n",
    "            if len(train_accs) > 3 and train_accs[-1] < train_accs[-3]:\n",
    "                break\n",
    "        \n",
    "        torch.save(self.encoder.state_dict(), f'Saved_models/Model_{model_name}val_acc_{int(val_accs[-1]*100)}encoder.pth')\n",
    "        torch.save(self.decoder.state_dict(), f'Saved_models/Model_{model_name}val_acc_{int(val_accs[-1]*100)}decoder.pth')\n",
    "\n",
    "    def eval(self, io_pairs):\n",
    "        correct = 0\n",
    "        results = []\n",
    "        for i in range(len(io_pairs)):\n",
    "            output, _ = self.predict(io_pairs[i][0])\n",
    "            results.append(output)\n",
    "            if output == io_pairs[i][1]:\n",
    "                correct += 1\n",
    "        return correct / len(io_pairs), results\n",
    "    \n",
    "    def predict(self, input_word, max_length=MAX_LENGTH):\n",
    "        with torch.no_grad():\n",
    "            input_tensor = word_to_tensor(self.input_lang, input_word)\n",
    "            input_length = input_tensor.size()[0]\n",
    "            encoder_hidden = self.encoder.initHidden()\n",
    "            encoder_cell = self.encoder.init_cell()\n",
    "            encoder_outputs = torch.zeros(max_length, self.encoder.hidden_size, device=device)\n",
    "\n",
    "            for ei in range(input_length):\n",
    "                encoder_output, encoder_hidden, encoder_cell = self.encoder(input_tensor[ei],\n",
    "                                                        encoder_hidden, encoder_cell)\n",
    "                encoder_outputs[ei] += encoder_output[0, 0]\n",
    "\n",
    "            decoder_input = torch.tensor([[SOS_token]], device=device)  # SOS\n",
    "\n",
    "            decoder_hidden = encoder_hidden[encoder_hidden.shape[0] - 1].unsqueeze(0)\n",
    "            decoder_cell = encoder_cell[encoder_cell.shape[0] - 1].unsqueeze(0)\n",
    "\n",
    "            decoded_word = ''\n",
    "            for i in range(self.decoder.num_layers - 1):\n",
    "                decoder_hidden = torch.cat((decoder_hidden, encoder_hidden[encoder_hidden.shape[0] - 1].unsqueeze(0)), 0)\n",
    "                decoder_cell = torch.cat((decoder_cell, encoder_cell[encoder_cell.shape[0] - 1].unsqueeze(0)), 0)\n",
    "\n",
    "            decoder_attentions = torch.zeros(max_length, max_length)\n",
    "\n",
    "            for di in range(max_length):\n",
    "                if self.attn:\n",
    "                    decoder_output, decoder_hidden, decoder_cell, decoder_attention = self.decoder(\n",
    "                    decoder_input, decoder_hidden, decoder_cell, encoder_outputs)\n",
    "                    decoder_attentions[di] = decoder_attention.data\n",
    "                else:\n",
    "                    decoder_output, decoder_hidden, decoder_cell = self.decoder(\n",
    "                    decoder_input, decoder_hidden, decoder_cell)\n",
    "                topv, topi = decoder_output.data.topk(1)\n",
    "                if topi.item() == EOS_token:\n",
    "                    # decoded_words.append('<EOS>')\n",
    "                    # decoded_word += '<EOS>'\n",
    "                    break\n",
    "                else:\n",
    "                    # decoded_words.append(output_lang.letter_index[topi.item()])\n",
    "                    decoded_word += self.output_lang.index_letter[topi.item()]\n",
    "                    # print(output_lang.index_letter[topi.item()])\n",
    "\n",
    "                decoder_input = topi.squeeze().detach()\n",
    "\n",
    "            return decoded_word, decoder_attentions[:di + 1]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch:  1  Loss:  4.873104095458984 Training Accuracy:  0.0 Validation Accuracy:  0.0\n"
     ]
    }
   ],
   "source": [
    "input_lang, output_lang, train_pairs = init_lang( 'eng','mar', type = 'train', reverse = False)\n",
    "validation_pairs = read_words('mar', type = 'valid')\n",
    "# print(input_lang.n_letters)\n",
    "encoder_hp = {\n",
    "    'input_size': input_lang.n_letters, \n",
    "    'embedding_size': 64, \n",
    "    'hidden_size': 512, \n",
    "    'num_layers': 2, \n",
    "    'dropout_p': 0.1,\n",
    "    'type': 'gru',\n",
    "    'bidirectional': False}\n",
    "decoder_hp = {\n",
    "    'hidden_size': 512, \n",
    "    'embedding_size': 64, \n",
    "    'output_size': output_lang.n_letters, \n",
    "    'num_layers': 2, \n",
    "    'dropout_p': 0.1,\n",
    "    'type': 'lstm',\n",
    "    'bidirectional': False}\n",
    "\n",
    "model = Transliterator(encoder_hp=encoder_hp, decoder_hp=decoder_hp, attn = True)\n",
    "# model_number +=1\n",
    "model.fit(train_pairs, validation_pairs, input_lang, output_lang, optimizer='adam', n_epochs = 1, learning_rate = 0.0003, teacher_forcing_ratio = 0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Create sweep with ID: yfqa6cy6\n",
      "Sweep URL: https://wandb.ai/cs20b004/CS6910_Assignment3/sweeps/yfqa6cy6\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: Agent Starting Run: hj9ljras with config:\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tbidirectional: False\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tdecoder_num_layers: 2\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tdecoder_type: lstm\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tdropout_p: 0\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tembedding_size: 64\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tencoder_num_layers: 2\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tencoder_type: lstm\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tepochs: 10\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \thidden_size: 384\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tlearning_rate: 0.0003\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \toptimizer: adam\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \tteacher_forcing_ratio: 0.7\n",
      "Failed to detect the name of this notebook, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable to enable code saving.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "wandb version 0.15.2 is available!  To upgrade, please run:\n",
       " $ pip install wandb --upgrade"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Tracking run with wandb version 0.13.11"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Run data is saved locally in <code>c:\\Users\\Aditya Patil\\Desktop\\IITM\\CS6910\\Assignment3\\wandb\\run-20230515_231306-hj9ljras</code>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Syncing run <strong><a href='https://wandb.ai/cs20b004/CS6910_Assignment3/runs/hj9ljras' target=\"_blank\">icy-sweep-1</a></strong> to <a href='https://wandb.ai/cs20b004/CS6910_Assignment3' target=\"_blank\">Weights & Biases</a> (<a href='https://wandb.me/run' target=\"_blank\">docs</a>)<br/>Sweep page: <a href='https://wandb.ai/cs20b004/CS6910_Assignment3/sweeps/yfqa6cy6' target=\"_blank\">https://wandb.ai/cs20b004/CS6910_Assignment3/sweeps/yfqa6cy6</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View project at <a href='https://wandb.ai/cs20b004/CS6910_Assignment3' target=\"_blank\">https://wandb.ai/cs20b004/CS6910_Assignment3</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View sweep at <a href='https://wandb.ai/cs20b004/CS6910_Assignment3/sweeps/yfqa6cy6' target=\"_blank\">https://wandb.ai/cs20b004/CS6910_Assignment3/sweeps/yfqa6cy6</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View run at <a href='https://wandb.ai/cs20b004/CS6910_Assignment3/runs/hj9ljras' target=\"_blank\">https://wandb.ai/cs20b004/CS6910_Assignment3/runs/hj9ljras</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch:  1  Loss:  1.2866709969610728 Training Accuracy:  0.26662109375 Validation Accuracy:  0.265380859375\n",
      "Epoch:  2  Loss:  0.6021577574973235 Training Accuracy:  0.408046875 Validation Accuracy:  0.3525390625\n",
      "Epoch:  3  Loss:  0.4685771653504404 Training Accuracy:  0.499453125 Validation Accuracy:  0.38671875\n",
      "Epoch:  4  Loss:  0.3911346473431443 Training Accuracy:  0.57654296875 Validation Accuracy:  0.42236328125\n",
      "Epoch:  5  Loss:  0.32988921786571174 Training Accuracy:  0.63494140625 Validation Accuracy:  0.417724609375\n",
      "Epoch:  6  Loss:  0.28371943828453905 Training Accuracy:  0.674921875 Validation Accuracy:  0.4248046875\n",
      "Epoch:  7  Loss:  0.24702618074901042 Training Accuracy:  0.715 Validation Accuracy:  0.433837890625\n",
      "Epoch:  8  Loss:  0.22001459658824818 Training Accuracy:  0.738046875 Validation Accuracy:  0.4306640625\n",
      "Epoch:  9  Loss:  0.19929555809721755 Training Accuracy:  0.759296875 Validation Accuracy:  0.435546875\n",
      "Epoch:  10  Loss:  0.17755590599565152 Training Accuracy:  0.78435546875 Validation Accuracy:  0.4326171875\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "Waiting for W&B process to finish... <strong style=\"color:green\">(success).</strong>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<style>\n",
       "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: left ; width: auto;} td:nth-child(2) {text-align: left ; width: 100%}\n",
       "    .wandb-row { display: flex; flex-direction: row; flex-wrap: wrap; justify-content: flex-start; width: 100% }\n",
       "    .wandb-col { display: flex; flex-direction: column; flex-basis: 100%; flex: 1; padding: 10px; }\n",
       "    </style>\n",
       "<div class=\"wandb-row\"><div class=\"wandb-col\"><h3>Run history:</h3><br/><table class=\"wandb\"><tr><td>Training Accuracy</td><td>▁▃▄▅▆▇▇▇██</td></tr><tr><td>Validation Accuracy</td><td>▁▅▆▇▇█████</td></tr><tr><td>epoch</td><td>▁▂▃▃▄▅▆▆▇█</td></tr><tr><td>train_loss</td><td>█▄▃▂▂▂▁▁▁▁</td></tr></table><br/></div><div class=\"wandb-col\"><h3>Run summary:</h3><br/><table class=\"wandb\"><tr><td>Training Accuracy</td><td>0.78436</td></tr><tr><td>Validation Accuracy</td><td>0.43262</td></tr><tr><td>epoch</td><td>10</td></tr><tr><td>train_loss</td><td>0.17756</td></tr></table><br/></div></div>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View run <strong style=\"color:#cdcd00\">icy-sweep-1</strong> at: <a href='https://wandb.ai/cs20b004/CS6910_Assignment3/runs/hj9ljras' target=\"_blank\">https://wandb.ai/cs20b004/CS6910_Assignment3/runs/hj9ljras</a><br/>Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Find logs at: <code>.\\wandb\\run-20230515_231306-hj9ljras\\logs</code>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: Sweep Agent: Waiting for job.\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: Sweep Agent: Exiting.\n"
     ]
    }
   ],
   "source": [
    "sweep_config = {\n",
    "    'name' : 'Overfixing',\n",
    "    'method': 'bayes', #grid, random\n",
    "    'metric': {\n",
    "        'name': 'Validation Accuracy',\n",
    "        'goal': 'maximize'\n",
    "    },\n",
    "    'parameters': {\n",
    "        'embedding_size': {\n",
    "            'values': [128, 192, 256]   }, \n",
    "        'hidden_size': {\n",
    "            'values': [192, 256]   },\n",
    "        'encoder_num_layers': { \n",
    "            'values': [1,2, 3]   },\n",
    "        'encoder_type': {'values' : ['lstm']},\n",
    "        'decoder_num_layers': {\n",
    "            'values': [1, 2, 3]   },\n",
    "        'decoder_type': {'values' : ['lstm']},\n",
    "        'learning_rate': {'values': [0.0006, 0.0005, 0.0003]},\n",
    "        'teacher_forcing_ratio': {'values': [ 0.6, 0.7]},\n",
    "        'optimizer' : {'values': ['adam']},\n",
    "        'epochs' : { 'values' : [10]},\n",
    "        'dropout_p' : {'values' : [0.0, 0.2]},\n",
    "        'bidirectional' : {'values' : [False]}\n",
    "    }\n",
    "}\n",
    "\n",
    "best_config = {\n",
    "    'name' : 'attention_testing',\n",
    "    'method': 'grid', #grid, random\n",
    "    'metric': {\n",
    "        'name': 'Validation Accuracy',\n",
    "        'goal': 'maximize'\n",
    "    },\n",
    "    'parameters': {\n",
    "        'embedding_size': \n",
    "            {'value': 64},\n",
    "        'hidden_size':\n",
    "            {'value': 384},\n",
    "        'encoder_num_layers':\n",
    "            {'value': 2},\n",
    "        'encoder_type':\n",
    "            {'value' : 'lstm'},\n",
    "        'decoder_num_layers':\n",
    "            {'value': 2},\n",
    "        'decoder_type':\n",
    "            {'value' : 'lstm'},\n",
    "        'learning_rate':    \n",
    "            {'value': 0.0003},\n",
    "        'teacher_forcing_ratio':\n",
    "            {'value': 0.7},\n",
    "        'optimizer' :\n",
    "            {'value': 'adam'},\n",
    "        'epochs' :\n",
    "            {'value' : 10},\n",
    "        'dropout_p' :\n",
    "            {'value' : 0.0},\n",
    "        'bidirectional' :\n",
    "            {'value' : False}\n",
    "    }\n",
    "}\n",
    "\n",
    "# sweep_id = wandb.sweep(sweep_config, project=\"CS6910_Assignment3\")\n",
    "sweep_id = wandb.sweep(sweep=best_config, project=\"CS6910_Assignment3\")\n",
    "\n",
    "def train():\n",
    "    config_defaults = {\n",
    "        'hidden_size': 256,\n",
    "        'encoder_num_layers': 1,\n",
    "        'encoder_type': 'gru',\n",
    "        'decoder_num_layers': 1,\n",
    "        'decoder_type': 'gru',\n",
    "        'learning_rate': 0.0001,\n",
    "        'teacher_forcing_ratio': 0.5,\n",
    "        'optimizer' : 'adam',\n",
    "        'epochs' : 5,\n",
    "        'dropout_p' : 0.1,\n",
    "        'bidirectional' : False\n",
    "    }\n",
    "\n",
    "    run = wandb.init()\n",
    "    config = wandb.config\n",
    "    name_str = f\"nle_{wandb.config['encoder_num_layers']}_nld_{wandb.config['decoder_num_layers']}_lr_{wandb.config['learning_rate']}_eu_{wandb.config['encoder_type']}_du_{wandb.config['decoder_type']}\"\n",
    "    wandb.run.name = name_str\n",
    "    model = Transliterator(\n",
    "        encoder_hp={\n",
    "            'input_size': input_lang.n_letters,\n",
    "            'embedding_size' : config.embedding_size, \n",
    "            'hidden_size': config.hidden_size, \n",
    "            'num_layers': config.encoder_num_layers, \n",
    "            'type': config.encoder_type,\n",
    "            'dropout_p' : config.dropout_p,\n",
    "            'bidirectional' : config.bidirectional\n",
    "            }, \n",
    "        decoder_hp={\n",
    "            'hidden_size': config.hidden_size, \n",
    "            'embedding_size': config.embedding_size,\n",
    "            'output_size': output_lang.n_letters, \n",
    "            'num_layers': config.decoder_num_layers, \n",
    "            'type': config.decoder_type,\n",
    "            'dropout_p' : config.dropout_p,\n",
    "            'bidirectional' : config.bidirectional\n",
    "            }, \n",
    "        attn = True)\n",
    "    model.fit(train_pairs, validation_pairs, input_lang, output_lang, n_epochs = config.epochs, learning_rate = config.learning_rate, teacher_forcing_ratio = config.teacher_forcing_ratio, use_wanb = True, model_name = \"Attn_on_best\")\n",
    "    run.finish()\n",
    "\n",
    "wandb.agent(sweep_id, train, count=2, project=\"CS6910_Assignment3\", entity=\"cs20b004\")\n",
    "wandb.finish()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "'dropout_p'",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "Cell \u001b[1;32mIn[12], line 13\u001b[0m\n\u001b[0;32m      1\u001b[0m encoder_hp \u001b[39m=\u001b[39m {\n\u001b[0;32m      2\u001b[0m     \u001b[39m'\u001b[39m\u001b[39minput_size\u001b[39m\u001b[39m'\u001b[39m: input_lang\u001b[39m.\u001b[39mn_letters,\n\u001b[0;32m      3\u001b[0m     \u001b[39m'\u001b[39m\u001b[39membedding_size\u001b[39m\u001b[39m'\u001b[39m : \u001b[39m64\u001b[39m,\n\u001b[0;32m      4\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mhidden_size\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m384\u001b[39m,\n\u001b[0;32m      5\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mnum_layers\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m1\u001b[39m,\n\u001b[0;32m      6\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mtype\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m'\u001b[39m\u001b[39mlstm\u001b[39m\u001b[39m'\u001b[39m}\n\u001b[0;32m      7\u001b[0m decoder_hp \u001b[39m=\u001b[39m {\n\u001b[0;32m      8\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mhidden_size\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m384\u001b[39m,\n\u001b[0;32m      9\u001b[0m     \u001b[39m'\u001b[39m\u001b[39membedding_size\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m64\u001b[39m,\n\u001b[0;32m     10\u001b[0m     \u001b[39m'\u001b[39m\u001b[39moutput_size\u001b[39m\u001b[39m'\u001b[39m: output_lang\u001b[39m.\u001b[39mn_letters,\n\u001b[0;32m     11\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mnum_layers\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m1\u001b[39m,\n\u001b[0;32m     12\u001b[0m     \u001b[39m'\u001b[39m\u001b[39mtype\u001b[39m\u001b[39m'\u001b[39m: \u001b[39m'\u001b[39m\u001b[39mgru\u001b[39m\u001b[39m'\u001b[39m}\n\u001b[1;32m---> 13\u001b[0m model \u001b[39m=\u001b[39m Transliterator(encoder_hp\u001b[39m=\u001b[39;49mencoder_hp, decoder_hp\u001b[39m=\u001b[39;49mdecoder_hp, attn \u001b[39m=\u001b[39;49m \u001b[39mFalse\u001b[39;49;00m)\n\u001b[0;32m     14\u001b[0m model\u001b[39m.\u001b[39mencoder\u001b[39m.\u001b[39mload_state_dict(torch\u001b[39m.\u001b[39mload(\u001b[39m'\u001b[39m\u001b[39mSaved_models/Best_encoder.pth\u001b[39m\u001b[39m'\u001b[39m))\n\u001b[0;32m     15\u001b[0m model\u001b[39m.\u001b[39mdecoder\u001b[39m.\u001b[39mload_state_dict(torch\u001b[39m.\u001b[39mload(\u001b[39m'\u001b[39m\u001b[39mSaved_models/Best_decoder.pth\u001b[39m\u001b[39m'\u001b[39m))\n",
      "Cell \u001b[1;32mIn[9], line 9\u001b[0m, in \u001b[0;36mTransliterator.__init__\u001b[1;34m(self, encoder_hp, decoder_hp, attn)\u001b[0m\n\u001b[0;32m      2\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m__init__\u001b[39m(\u001b[39mself\u001b[39m, encoder_hp, decoder_hp, attn \u001b[39m=\u001b[39m \u001b[39mTrue\u001b[39;00m):\n\u001b[0;32m      3\u001b[0m     \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mencoder \u001b[39m=\u001b[39m EncoderRNN(\n\u001b[0;32m      4\u001b[0m                     encoder_hp[\u001b[39m'\u001b[39m\u001b[39minput_size\u001b[39m\u001b[39m'\u001b[39m],\n\u001b[0;32m      5\u001b[0m                     encoder_hp[\u001b[39m'\u001b[39m\u001b[39membedding_size\u001b[39m\u001b[39m'\u001b[39m],\n\u001b[0;32m      6\u001b[0m                     encoder_hp[\u001b[39m'\u001b[39m\u001b[39mhidden_size\u001b[39m\u001b[39m'\u001b[39m], \n\u001b[0;32m      7\u001b[0m                     encoder_hp[\u001b[39m'\u001b[39m\u001b[39mnum_layers\u001b[39m\u001b[39m'\u001b[39m], \n\u001b[0;32m      8\u001b[0m                     \u001b[39mtype\u001b[39m \u001b[39m=\u001b[39m encoder_hp[\u001b[39m'\u001b[39m\u001b[39mtype\u001b[39m\u001b[39m'\u001b[39m],\n\u001b[1;32m----> 9\u001b[0m                     dropout_p\u001b[39m=\u001b[39m encoder_hp[\u001b[39m'\u001b[39;49m\u001b[39mdropout_p\u001b[39;49m\u001b[39m'\u001b[39;49m],\n\u001b[0;32m     10\u001b[0m                     bidirectional \u001b[39m=\u001b[39m encoder_hp[\u001b[39m'\u001b[39m\u001b[39mbidirectional\u001b[39m\u001b[39m'\u001b[39m]\n\u001b[0;32m     11\u001b[0m                     )\u001b[39m.\u001b[39mto(device)\n\u001b[0;32m     12\u001b[0m     \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mattn \u001b[39m=\u001b[39m attn\n\u001b[0;32m     13\u001b[0m     \u001b[39mif\u001b[39;00m attn:\n",
      "\u001b[1;31mKeyError\u001b[0m: 'dropout_p'"
     ]
    }
   ],
   "source": [
    "encoder_hp = {\n",
    "    'input_size': input_lang.n_letters,\n",
    "    'embedding_size' : 64,\n",
    "    'hidden_size': 384,\n",
    "    'num_layers': 1,\n",
    "    'type': 'lstm',\n",
    "    'dropout_p' : 0.0,\n",
    "    bidirectional' : False}\n",
    "decoder_hp = {\n",
    "    'hidden_size': 384,\n",
    "    'embedding_size': 64,\n",
    "    'output_size': output_lang.n_letters,\n",
    "    'num_layers': 1,\n",
    "    'type': 'gru'}\n",
    "model = Transliterator(encoder_hp=encoder_hp, decoder_hp=decoder_hp, attn = False)\n",
    "model.encoder.load_state_dict(torch.load('Saved_models/Best_encoder.pth'))\n",
    "model.decoder.load_state_dict(torch.load('Saved_models/Best_decoder.pth'))\n",
    "model.input_lang = input_lang\n",
    "model.output_lang = output_lang\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "#LOAD BEST MODEL\n",
    "encoder_hp_1= {\n",
    "    'input_size': input_lang.n_letters,\n",
    "    'embedding_size' : 64,\n",
    "    'hidden_size': 384,\n",
    "    'num_layers': 2,\n",
    "    'type': 'lstm',\n",
    "    'dropout_p' : 0,\n",
    "    'bidirectional' : False\n",
    "    }\n",
    "decoder_hp_1 = {\n",
    "    'hidden_size': 384,\n",
    "    'embedding_size': 64,\n",
    "    'output_size': output_lang.n_letters,\n",
    "    'num_layers': 2,\n",
    "    'dropout_p' : 0,\n",
    "    'bidirectional' : False,\n",
    "    'type': 'lstm'}\n",
    "\n",
    "best_m = Transliterator(encoder_hp=encoder_hp_1, decoder_hp=decoder_hp_1, attn = False)\n",
    "best_m.encoder.load_state_dict(torch.load('Saved_models/val_acc_44encoder.pth'))\n",
    "best_m.decoder.load_state_dict(torch.load('Saved_models/val_acc_44decoder.pth'))\n",
    "best_m.input_lang = input_lang\n",
    "best_m.output_lang = output_lang\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "#BEST MODEL WITH ATTENTION\n",
    "best_m_attn = Transliterator(encoder_hp=encoder_hp_1, decoder_hp=decoder_hp_1, attn = True)\n",
    "best_m_attn.encoder.load_state_dict(torch.load('Saved_models/Model_Attn_on_bestval_acc_43encoder.pth'))\n",
    "best_m_attn.decoder.load_state_dict(torch.load('Saved_models/Model_Attn_on_bestval_acc_43decoder.pth'))\n",
    "best_m_attn.input_lang = input_lang\n",
    "best_m_attn.output_lang = output_lang"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_pairs = read_words('mar', type = 'test')\n",
    "# acc, output_test = best_m.eval(test_pairs)\n",
    "#store in csv original and predicted\n",
    "acc_attn,output_test_attn = best_m_attn.eval(test_pairs)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "df = pd.DataFrame(list(zip([p[0] for p in test_pairs],[p[1] for p in test_pairs], output_test_attn)), columns = ['Original','Translated', 'Predicted'])\n",
    "df.to_csv('predictions_attention.csv', index = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_inputs = [\n",
    "'heetler',\n",
    "'kaushik',\n",
    "'bombay',\n",
    "'loid',\n",
    "'shikshanvishayak',\n",
    "'purushottam',\n",
    "'shrivas',\n",
    "'palale',\n",
    "'baabhlichi',\n",
    "'pitali']\n",
    "\n",
    "\n",
    "attn_outputs = []\n",
    "attn_heatmaps = []\n",
    "for inp in test_inputs:\n",
    "    output, _ = best_m_attn.predict(inp)\n",
    "    attn_outputs.append(output)\n",
    "    attn_heatmaps.append(_)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.font_manager as fm\n",
    "def plot_attention_heatmap(word,translated_word,attn_heatmap, ax,i, font_path=\"Lohit-Devanagari.ttf\"):\n",
    "\n",
    "    input_word_len = len(word)\n",
    "    output_word_len = len(translated_word)\n",
    "\n",
    "    ax.imshow(attn_heatmap[:, :input_word_len])\n",
    "\n",
    "    font_prop = fm.FontProperties(fname=font_path, size=18)\n",
    "\n",
    "    ax.set_xticks(np.arange(input_word_len))\n",
    "    ax.set_yticks(np.arange(output_word_len))\n",
    "\n",
    "    ax.set_xticklabels(list(word))\n",
    "    ax.set_yticklabels(list(translated_word), fontproperties=font_prop)\n",
    "    ax.set_xlabel(\"Input Word\")\n",
    "    ax.set_ylabel(\"Translated Word\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 2000x1000 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "#get 1 images and plot\n",
    "fig, ax = plt.subplots(2, 5, figsize=(20, 10))\n",
    "\n",
    "for i in range(2):\n",
    "    for j in range(5):\n",
    "        plot_attention_heatmap(test_inputs[i*5 + j],attn_outputs[i*5 + j],attn_heatmaps[i*5 + j], ax[i][j], str(i*5 + j))\n",
    "\n",
    "# for i in range(10):\n",
    "# # plot_attention_heatmap(test_inputs[0],attn_outputs[0],attn_heatmaps[0], ax, 0)\n",
    "#   plot_attention_heatmap(test_inputs[i],attn_outputs[i],attn_heatmaps[i], ax[i], str(i))  "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.4"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}