Pytorch_NLP_tutorial.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "2d244895",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import os,sys\n",
    "import imageio\n",
    "import random\n",
    "from tqdm import tqdm\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.manifold import TSNE\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import torchtext\n",
    "\n",
    "\n",
    "SEED = 1\n",
    "torch.manual_seed(SEED)\n",
    "torch.backends.cudnn.deterministic = True"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "290525c7",
   "metadata": {},
   "source": [
    "# Computing Word Embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "4393145c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[(['When', 'forty'], 'winters'), (['forty', 'winters'], 'shall'), (['winters', 'shall'], 'besiege')]\n",
      "[521.4706325531006, 519.0597877502441, 516.6645703315735, 514.2839488983154, 511.91663551330566, 509.56114864349365, 507.216548204422, 504.88212490081787, 502.5561282634735, 500.2394006252289, 497.9295918941498, 495.6279344558716, 493.3345820903778, 491.047593832016, 488.76683616638184, 486.49101734161377, 484.2214653491974, 481.95589447021484, 479.6933813095093, 477.4305477142334, 475.16832160949707, 472.9073646068573, 470.64716625213623, 468.38526368141174, 466.12247824668884, 463.85905861854553, 461.594797372818, 459.3299341201782, 457.0623903274536, 454.7914574146271, 452.5185809135437, 450.2426450252533, 447.96203899383545, 445.6770007610321, 443.386212348938, 441.08977484703064, 438.7861657142639, 436.4769084453583, 434.16180741786957, 431.8409335613251, 429.5134210586548, 427.1798758506775, 424.83789336681366, 422.48745810985565, 420.128090262413, 417.76121282577515, 415.385670542717, 413.0027425289154, 410.61292135715485, 408.21440291404724]\n",
      "tensor([-0.7882,  0.3171,  0.8264,  0.8604,  0.4969,  0.4431, -0.8861,  0.5304,\n",
      "         0.2736,  0.3438], grad_fn=<SelectBackward>)\n"
     ]
    }
   ],
   "source": [
    "CONTEXT_SIZE = 2\n",
    "EMBEDDING_DIM = 10\n",
    "# We will use Shakespeare Sonnet 2\n",
    "test_sentence = \"\"\"When forty winters shall besiege thy brow,\n",
    "And dig deep trenches in thy beauty's field,\n",
    "Thy youth's proud livery so gazed on now,\n",
    "Will be a totter'd weed of small worth held:\n",
    "Then being asked, where all thy beauty lies,\n",
    "Where all the treasure of thy lusty days;\n",
    "To say, within thine own deep sunken eyes,\n",
    "Were an all-eating shame, and thriftless praise.\n",
    "How much more praise deserv'd thy beauty's use,\n",
    "If thou couldst answer 'This fair child of mine\n",
    "Shall sum my count, and make my old excuse,'\n",
    "Proving his beauty by succession thine!\n",
    "This were to be new made when thou art old,\n",
    "And see thy blood warm when thou feel'st it cold.\"\"\".split()\n",
    "# we should tokenize the input, but we will ignore that for now\n",
    "# build a list of tuples.  Each tuple is ([ word_i-2, word_i-1 ], target word)\n",
    "trigrams = [([test_sentence[i], test_sentence[i + 1]], test_sentence[i + 2])\n",
    "            for i in range(len(test_sentence) - 2)]\n",
    "# print the first 3, just so you can see what they look like\n",
    "print(trigrams[:3])\n",
    "\n",
    "vocab = set(test_sentence)\n",
    "word_to_ix = {word: i for i, word in enumerate(vocab)}\n",
    "\n",
    "\n",
    "class NGramLanguageModeler(nn.Module):\n",
    "\n",
    "    def __init__(self, vocab_size, embedding_dim, context_size):\n",
    "        super(NGramLanguageModeler, self).__init__()\n",
    "        self.embeddings = nn.Embedding(vocab_size, embedding_dim)\n",
    "        self.linear1 = nn.Linear(context_size * embedding_dim, 128)\n",
    "        self.linear2 = nn.Linear(128, vocab_size)\n",
    "\n",
    "    def forward(self, inputs):\n",
    "        embeds = self.embeddings(inputs).view((1, -1))\n",
    "        out = F.relu(self.linear1(embeds))\n",
    "        out = self.linear2(out)\n",
    "        log_probs = F.log_softmax(out, dim=1)\n",
    "        return log_probs\n",
    "\n",
    "\n",
    "losses = []\n",
    "loss_function = nn.NLLLoss()\n",
    "model = NGramLanguageModeler(len(vocab), EMBEDDING_DIM, CONTEXT_SIZE)\n",
    "optimizer = optim.SGD(model.parameters(), lr=0.001)\n",
    "\n",
    "for epoch in range(50):\n",
    "    total_loss = 0\n",
    "    for context, target in trigrams:\n",
    "\n",
    "        # Step 1. Prepare the inputs to be passed to the model (i.e, turn the words\n",
    "        # into integer indices and wrap them in tensors)\n",
    "        context_idxs = torch.tensor([word_to_ix[w] for w in context], dtype=torch.long)\n",
    "\n",
    "        # Step 2. Recall that torch *accumulates* gradients. Before passing in a\n",
    "        # new instance, you need to zero out the gradients from the old\n",
    "        # instance\n",
    "        model.zero_grad()\n",
    "\n",
    "        # Step 3. Run the forward pass, getting log probabilities over next\n",
    "        # words\n",
    "        log_probs = model(context_idxs)\n",
    "\n",
    "        # Step 4. Compute your loss function. (Again, Torch wants the target\n",
    "        # word wrapped in a tensor)\n",
    "        loss = loss_function(log_probs, torch.tensor([word_to_ix[target]], dtype=torch.long))\n",
    "\n",
    "        # Step 5. Do the backward pass and update the gradient\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        # Get the Python number from a 1-element Tensor by calling tensor.item()\n",
    "        total_loss += loss.item()\n",
    "    losses.append(total_loss)\n",
    "print(losses)  # The loss decreased every iteration over the training data!\n",
    "\n",
    "# To get the embedding of a particular word, e.g. \"beauty\"\n",
    "print(model.embeddings.weight[word_to_ix[\"beauty\"]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "494ca064",
   "metadata": {},
   "outputs": [],
   "source": [
    "# TSNE visualization of embeddings\n",
    "\n",
    "embs = torch.stack([model.embeddings.weight[word_to_ix[word]] for word in vocab]).detach().numpy()\n",
    "labels = [word for word in vocab]\n",
    "\n",
    "embs_2d = TSNE(n_components=2).fit_transform(embs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a8d73e4d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=(8,8))\n",
    "ax.scatter(embs_2d[:,0], embs_2d[:,1])\n",
    "\n",
    "for i, txt in enumerate(labels):\n",
    "    ax.annotate(txt, (embs_2d[i,0], embs_2d[i,1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "37c571b5",
   "metadata": {},
   "source": [
    "# Sentiment Classification on the IMDB dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "07244759",
   "metadata": {},
   "outputs": [],
   "source": [
    "TEXT = torchtext.data.Field(tokenizer_language = 'basic_english')\n",
    "LABEL = torchtext.data.LabelField(dtype = torch.float)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "d924cb4f",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data, test_data = torchtext.datasets.IMDB.splits(TEXT, LABEL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "724d422f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of training examples: 25000\n",
      "Number of testing examples: 25000\n"
     ]
    }
   ],
   "source": [
    "print(f'Number of training examples: {len(train_data)}')\n",
    "print(f'Number of testing examples: {len(test_data)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "bea0c19e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'text': ['We', 'do', 'not', 'come', 'across', 'movies', 'on', 'brother-sister', 'relationship', 'in', 'Indian', 'cinema,', 'or', 'any', 'other', 'language', 'or', 'medium.', 'This', 'relationship', 'has', 'several', 'aspects', 'which', 'have', 'not', 'been', 'exploited', 'in', 'movies', 'or', 'novels.', 'Typically,', 'a', 'sister', 'is', 'depicted', 'as', 'a', 'pile-on', 'who', 'can', 'be', 'used', 'for', 'ransom', 'in', 'the', 'climax.', 'This', 'movie', 'treats', 'the', 'subject', 'in', 'an', 'entirely', 'different', 'light.<br', '/><br', '/>It', 'is', 'inspired', 'by', 'George', \"Eliot's\", 'novel', '\"The', 'Mill', 'on', 'the', 'Floss\".', 'The', 'brother', 'is', 'very', 'prosaic,', 'all-good,', 'the', 'blue-eyed', 'boy', 'who', 'is', 'a', 'conventionally', 'good', 'son', 'and', 'a', 'favorite', 'with', 'his', 'mother.', 'The', 'sister', 'is', 'romantic,', 'wild', 'and', 'defiant', 'of', 'the', 'unwritten', 'rules', 'of', 'the', 'society.', 'In', 'spite', 'of', 'this,', 'the', 'love', 'of', 'the', 'brother-sister', 'is', 'the', 'winner.<br', '/><br', '/>This', 'movie', 'is', 'about', 'the', 'love', 'of', 'the', 'two', 'siblings', 'who', 'are', 'separated', 'in', 'childhood', 'and', 'revival', 'of', 'the', 'same', 'feeling', 'when', 'they', 'meet', 'years', 'later.', 'It', 'is', 'also', 'the', 'quest', 'of', 'the', 'subdued', 'brother', 'to', 'reunite', 'with', 'his', 'sister', 'who', 'has', 'chosen', 'to', 'be', 'wild', 'to', 'defy', 'the', 'world.<br', '/><br', '/>Although', 'the', 'movie', 'and', 'the', 'novel', 'are', 'set', 'about', '3', 'centuries', 'apart', 'in', 'two', 'distant', 'countries,', 'yet', 'the', 'sentiments', 'are', 'the', 'same', 'and', 'still', 'hold', 'true.'], 'label': 'pos'}\n"
     ]
    }
   ],
   "source": [
    "print(vars(train_data.examples[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "2cf78334",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data, valid_data = train_data.split(random_state = random.seed(SEED))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "4a3aa395",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of training examples: 17500\n",
      "Number of validation examples: 7500\n",
      "Number of testing examples: 25000\n"
     ]
    }
   ],
   "source": [
    "print(f'Number of training examples: {len(train_data)}')\n",
    "print(f'Number of validation examples: {len(valid_data)}')\n",
    "print(f'Number of testing examples: {len(test_data)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "4e7c9868",
   "metadata": {},
   "outputs": [],
   "source": [
    "TEXT.build_vocab(train_data, vectors='glove.6B.50d')\n",
    "LABEL.build_vocab(train_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "00e475de",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Unique tokens in TEXT vocabulary: 224776\n",
      "Unique tokens in LABEL vocabulary: 2\n"
     ]
    }
   ],
   "source": [
    "print(f\"Unique tokens in TEXT vocabulary: {len(TEXT.vocab)}\")\n",
    "print(f\"Unique tokens in LABEL vocabulary: {len(LABEL.vocab)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "0378de67",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[('the', 201300), ('a', 109208), ('and', 107232), ('of', 100305), ('to', 92934), ('is', 72631), ('in', 59955), ('I', 46058), ('that', 45393), ('this', 40088), ('it', 37880), ('/><br', 35840), ('was', 32602), ('as', 30054), ('with', 29360), ('for', 28760), ('but', 23528), ('The', 23463), ('on', 21516), ('movie', 21316)]\n"
     ]
    }
   ],
   "source": [
    "print(TEXT.vocab.freqs.most_common(20))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "28f48635",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_batch_sentence_embeddings(batch):\n",
    "    sentence_embeddings = []\n",
    "    for text in batch.text.transpose(1,0):\n",
    "        filtered_text = text[(text!=0)&(text!=1)] # removes <unk> and <pad> tokens\n",
    "        sentence_emb = torch.mean(TEXT.vocab.vectors[filtered_text],axis=0)\n",
    "        sentence_embeddings.append(sentence_emb)\n",
    "    \n",
    "    return torch.stack(sentence_embeddings)\n",
    "\n",
    "class LogisticRegression(torch.nn.Module):\n",
    "    def __init__(self, input_dim):\n",
    "        super(LogisticRegression, self).__init__()\n",
    "        self.linear = torch.nn.Linear(input_dim,1)\n",
    "\n",
    "    def forward(self, x):\n",
    "        outputs = self.linear(x)\n",
    "        return outputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "abc56c94",
   "metadata": {},
   "outputs": [],
   "source": [
    "# dataloader\n",
    "BATCH_SIZE = 64\n",
    "\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "\n",
    "train_iterator, valid_iterator, test_iterator = torchtext.data.BucketIterator.splits(\n",
    "    (train_data, valid_data, test_data), \n",
    "    batch_size = BATCH_SIZE,\n",
    "    device = device)\n",
    "\n",
    "\n",
    "# model + loss + optimizer\n",
    "model = LogisticRegression(input_dim=50)\n",
    "criterion = torch.nn.BCEWithLogitsLoss()\n",
    "optimizer = torch.optim.SGD(model.parameters(), lr=1e-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "ff3bd0c9",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "274it [00:05, 45.76it/s]\n",
      "274it [00:05, 45.77it/s]\n",
      "274it [00:05, 46.96it/s]\n",
      "274it [00:06, 45.52it/s]\n",
      "274it [00:05, 46.35it/s]\n",
      "274it [00:05, 46.23it/s]\n",
      "274it [00:06, 44.65it/s]\n",
      "274it [00:05, 45.89it/s]\n",
      "274it [00:06, 45.05it/s]\n",
      "274it [00:06, 45.64it/s]\n",
      "274it [00:05, 47.38it/s]\n",
      "274it [00:05, 46.55it/s]\n",
      "274it [00:05, 46.61it/s]\n",
      "274it [00:05, 46.19it/s]\n",
      "274it [00:05, 46.89it/s]\n",
      "274it [00:05, 46.94it/s]\n",
      "274it [00:05, 46.94it/s]\n",
      "274it [00:05, 46.66it/s]\n",
      "274it [00:06, 45.65it/s]\n",
      "274it [00:06, 43.37it/s]\n",
      "274it [00:05, 45.76it/s]\n",
      "274it [00:06, 45.58it/s]\n",
      "274it [00:06, 43.60it/s]\n",
      "274it [00:05, 47.07it/s]\n",
      "274it [00:05, 46.28it/s]\n",
      "274it [00:05, 47.80it/s]\n",
      "274it [00:05, 46.78it/s]\n",
      "274it [00:05, 47.22it/s]\n",
      "274it [00:05, 47.32it/s]\n",
      "274it [00:05, 45.93it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.6913998126983643, 0.6832635402679443, 0.6864414215087891, 0.6794276237487793, 0.6841809749603271, 0.6758760213851929, 0.6698373556137085, 0.673502504825592, 0.6794764399528503, 0.6841993927955627, 0.6747421622276306, 0.6695747375488281, 0.6699034571647644, 0.6656400561332703, 0.6751526594161987, 0.6630975008010864, 0.6673204302787781, 0.6756502389907837, 0.6629698276519775, 0.675130307674408, 0.6662340760231018, 0.659463107585907, 0.6653755307197571, 0.6802966594696045, 0.6643118858337402, 0.6481318473815918, 0.6546914577484131, 0.6355668902397156, 0.6521721482276917, 0.6508311033248901]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "epochs = 30\n",
    "\n",
    "losses = []\n",
    "for epoch in range(int(epochs)):\n",
    "    for i,batch in tqdm(enumerate(train_iterator)):\n",
    "        \n",
    "        sentence_embeddings = get_batch_sentence_embeddings(batch)\n",
    "        optimizer.zero_grad()\n",
    "        outputs = model(sentence_embeddings)\n",
    "        loss = criterion(outputs.reshape(-1), batch.label)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "    losses.append(loss.item())\n",
    "\n",
    "print(losses)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "97873a02",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_metrics(iterator,model):\n",
    "    model.eval()\n",
    "    all_preds = []\n",
    "    all_labels = []\n",
    "    for batch in iterator:\n",
    "        sentence_embeddings = get_batch_sentence_embeddings(batch)\n",
    "        with torch.no_grad():\n",
    "            all_preds.extend(list(model(sentence_embeddings).detach().numpy()))\n",
    "        all_labels.extend(list(batch.label.detach().numpy()))\n",
    "    \n",
    "    return {'auc': roc_auc_score(all_labels,all_preds)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "778252d3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'auc': 0.7285463104000001}"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compute_metrics(test_iterator,model)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b4b2551b",
   "metadata": {},
   "source": [
    "## References"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f8bda2b7",
   "metadata": {},
   "source": [
    "Guthrie, Robert. “Word Embeddings: Encoding Lexical Semantics¶.” Word Embeddings: Encoding Lexical Semantics - PyTorch Tutorials 1.8.1+cu102 Documentation, pytorch.org/tutorials/beginner/nlp/word_embeddings_tutorial.html#sphx-glr-beginner-nlp-word-embeddings-tutorial-py.\n",
    "\n",
    "Trevett, Ben. “Bentrevett/Pytorch-Sentiment-Analysis.” GitHub, github.com/bentrevett/pytorch-sentiment-analysis. "
   ]
  }
 ],
 "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": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}