Lecture 10 - CNN for text classification.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Lecture 10 - CNN for text classification.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5EolzpPTBU2C"
      },
      "source": [
        "# Case Study: CNN for text classification\n",
        "\n",
        "\n",
        "\n",
        "To illustrate the process of re-implmenting a pipeline\n",
        "\n",
        "[Large Movie Review Dataset](https://ai.stanford.edu/~amaas/data/sentiment/)\n",
        "\n",
        "Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts. (2011). \"*Learning Word Vectors for Sentiment Analysis.*\" The 49th Annual Meeting of the Association for Computational Linguistics (ACL 2011).\n",
        "\n",
        "Yoon Kim. \"[*Convolutional Neural Networks for Sentence Classification*](https://arxiv.org/abs/1408.5882).\" arXiv. 2014.\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Q1U1kcW1BC9D",
        "outputId": "4fe6e2fd-885e-47d3-e145-7996c84a9a38"
      },
      "source": [
        "import os.path\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "from tqdm import tqdm\n",
        "import numpy as np\n",
        "\n",
        "if torch.cuda.is_available():\n",
        "  device = torch.device(\"cuda\")\n",
        "  for i in range(torch.cuda.device_count()):\n",
        "    print(torch.cuda.get_device_name(i))\n",
        "else:\n",
        "  device = torch.device(\"cpu\")\n",
        "print(\"Device:\", device)"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tesla T4\n",
            "Device: cuda\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HHUqUIsXcN0C"
      },
      "source": [
        "Loading the data set"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GzGTee0-_ous",
        "outputId": "51d52264-c4a3-4db8-ade8-fb8c718ffbe2"
      },
      "source": [
        "!wget -N https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\n",
        "!tar zxf aclImdb_v1.tar.gz --skip-old-files\n",
        "!ls aclImdb/"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-05-11 13:57:36--  https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\n",
            "Resolving ai.stanford.edu (ai.stanford.edu)... 171.64.68.10\n",
            "Connecting to ai.stanford.edu (ai.stanford.edu)|171.64.68.10|:443... connected.\n",
            "HTTP request sent, awaiting response... 304 Not Modified\n",
            "File ‘aclImdb_v1.tar.gz’ not modified on server. Omitting download.\n",
            "\n",
            "imdbEr.txt  imdb.vocab\tREADME\ttest  train\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RarqYUzd6kcH",
        "outputId": "8862f4b9-4028-4b7c-d08c-916d130b4542"
      },
      "source": [
        "class PolarityDataset(torch.utils.data.Dataset):\n",
        "  def __init__(self, positive_dir, negative_dir, transform=None):\n",
        "    assert os.path.exists(positive_dir) and os.path.isdir(positive_dir)\n",
        "    self._sourcefiles = [os.path.join(positive_dir, fn) for fn in os.listdir(positive_dir) if fn.endswith(\".txt\")]\n",
        "    self._labels = [1]*len(self._sourcefiles)\n",
        "    assert os.path.exists(negative_dir) and os.path.isdir(negative_dir)\n",
        "    self._sourcefiles.extend([os.path.join(negative_dir, fn) for fn in os.listdir(negative_dir) if fn.endswith(\".txt\")])\n",
        "    self._labels.extend([0]*(len(self._sourcefiles)-len(self._labels)))\n",
        "    for fn in self._sourcefiles:\n",
        "      assert os.path.exists(fn), \"%s is missing\" % fn\n",
        "    self._transform = transform\n",
        "\n",
        "  def __len__(self):\n",
        "    return len(self._sourcefiles)\n",
        "\n",
        "  def __getitem__(self, idx):\n",
        "    if torch.is_tensor(idx):\n",
        "      idx = idx.tolist()\n",
        "    with open(self._sourcefiles[idx], 'r') as f:\n",
        "      ret = {'text': f.read(), \n",
        "             'label': self._labels[idx]}\n",
        "    if self._transform:\n",
        "      ret = self._transform(ret)\n",
        "    return ret\n",
        "\n",
        "training_data = PolarityDataset(positive_dir=\"aclImdb/train/pos/\", \n",
        "                                negative_dir=\"aclImdb/train/neg/\")\n",
        "\n",
        "for i in [0, 1, 2, 24000, 24001, 24002]:\n",
        "  print(training_data[i]['label'], \":\", training_data[i]['text'][:100])"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1 : Well... easily my favourite TV series ever. Call me a walking mail cliché but include violence, mafi\n",
            "1 : Yes, this gets the full ten stars. It's plain as day that this fill is genius. The universe sent Tre\n",
            "1 : I saw this movie when I was about 8-years-old and I liked it but it wasn't until I watched it again \n",
            "0 : I loved Dewaere in Series Noir. His talent is trivialized in \"The Waltzers\" aka \"Going Places\". Okay\n",
            "0 : This is one of those movies that's trying to be moody and tense, and instead, ends up tripping all o\n",
            "0 : Let me just say I loved the original Boogeyman. Sure, it's a flawed clichéd 80s horror movie, but he\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9BYFgGwjdJEN"
      },
      "source": [
        "Create vocabulary"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sCRyRnCPdJKG",
        "outputId": "800ea50a-2cb6-42f3-ee3b-007c4843847a"
      },
      "source": [
        "from nltk.tokenize import wordpunct_tokenize\n",
        "\n",
        "class Tokenize(object):\n",
        "    def __call__(self, data):\n",
        "        text, label = data['text'], data['label']\n",
        "        return {'text': [word.lower() for word in wordpunct_tokenize(text)],\n",
        "                'label': label}\n",
        "\n",
        "training_data = PolarityDataset(positive_dir=\"aclImdb/train/pos/\", \n",
        "                                negative_dir=\"aclImdb/train/neg/\",\n",
        "                                transform=Tokenize())\n",
        "test_data = PolarityDataset(positive_dir=\"aclImdb/test/pos/\", \n",
        "                            negative_dir=\"aclImdb/test/neg/\",\n",
        "                            transform=Tokenize())\n",
        "\n",
        "print(\"Loading training data...\", end=\"\")\n",
        "dataset_vocab = set()\n",
        "for sample in training_data:\n",
        "  sample_tokens = set(sample['text'])\n",
        "  dataset_vocab.update(sample_tokens)\n",
        "print(\"done\")\n",
        "\n",
        "print(\"Loading test data...\", end=\"\")\n",
        "for sample in test_data:\n",
        "  sample_tokens = set(sample['text'])\n",
        "  dataset_vocab.update(sample_tokens)\n",
        "print(\"done\")\n",
        "\n",
        "print(\"Found %i tokens in the dataset\" % len(dataset_vocab))"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading training data...done\n",
            "Loading test data...done\n",
            "Found 104339 tokens in the dataset\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xY73LU6XbFxI"
      },
      "source": [
        "## Word embedding\n",
        "\n",
        "We need a word embedding"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yMShKdhDbGJK",
        "outputId": "c1637da6-fed2-4c18-d447-5548f5bcbb6c"
      },
      "source": [
        "from google.colab import drive\n",
        "\n",
        "drive.mount('/content/drive')\n",
        "base = \"\"\"/content/drive/MyDrive/GloVe\"\"\"\n",
        "assert os.path.exists(base)\n",
        "!mkdir .vector_cache\n",
        "# !cp /content/drive/MyDrive/GloVe/glove.6B.zip .vector_cache\n",
        "!ln -s /content/drive/MyDrive/GloVe/glove.6B.zip .vector_cache/"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n",
            "mkdir: cannot create directory ‘.vector_cache’: File exists\n",
            "ln: failed to create symbolic link '.vector_cache/glove.6B.zip': File exists\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3pv7EtNybCTE"
      },
      "source": [
        "from torchtext.vocab import GloVe\n",
        "glove = GloVe(name='6B', dim=100)"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 589
        },
        "id": "gglp6ovoaofn",
        "outputId": "1465afb5-6dde-4753-cfb4-6a6605fdcfa5"
      },
      "source": [
        "print(\"%i words in GloVe\" % len(glove.itos))\n",
        "print(\"%i words in the dataset\" % len(dataset_vocab))\n",
        "print(\"%i overlapping words\" % len(dataset_vocab.intersection(set(glove.itos))))\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "from scipy.optimize import minimize_scalar\n",
        "\n",
        "a = np.linspace(0, 2*np.pi, 100)\n",
        "r1 = np.sqrt(len(glove.itos)/np.pi)\n",
        "r2 = np.sqrt(len(dataset_vocab)/np.pi)\n",
        "A = np.random.uniform(-r2, r2, size=(int(1000*(4/np.pi)), 2))\n",
        "A = A[np.sqrt(np.sum(A**2, axis=1))<=r2, :]\n",
        "p  = len(dataset_vocab.intersection(set(glove.itos)))/len(dataset_vocab)\n",
        "def f(d):\n",
        "  B = A.copy()\n",
        "  B[:, 1] += d\n",
        "  return np.abs(p-np.sum(np.sqrt(np.sum(B**2, axis=1))<=r1)/B.shape[0])\n",
        "d = minimize_scalar(f, bounds=(r1-r2, r1+r2), method='bounded').x\n",
        "\n",
        "plt.figure(figsize=(4, 4), dpi=150)\n",
        "plt.plot(r1*np.cos(a), r1*np.sin(a))\n",
        "plt.text(0, 0, \"GloVe\", fontsize=14, horizontalalignment='center', verticalalignment='center')\n",
        "plt.plot(d+r2*np.cos(a), r2*np.sin(a))\n",
        "plt.text(d, 0, \"Dataset\", fontsize=14, horizontalalignment='center', verticalalignment='center')\n",
        "# A[:, 0] += d\n",
        "# plt.scatter(A[:, 0], A[:, 1], s=3, c=np.sqrt(np.sum(A**2, axis=1))<=r1, marker='.', cmap='tab10')\n",
        "plt.title(\"Vocabularies\")\n",
        "plt.xticks([])\n",
        "plt.yticks([])\n",
        "plt.axis('equal')\n",
        "plt.show()"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "400000 words in GloVe\n",
            "104339 words in the dataset\n",
            "75862 overlapping words\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 600x600 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "A-mFGsHydG3A"
      },
      "source": [
        "The dataset vocabulary does not overlap completely with the embedding vocabulary. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4nEq99_zjdTD",
        "outputId": "07188dde-b042-4ab7-9f50-ee0aa2456d01"
      },
      "source": [
        "embedding_vocab = dataset_vocab.intersection(set(glove.itos))\n",
        "\n",
        "idx2token = ['<UNK>', '<PAD>']\n",
        "idx2token.extend(embedding_vocab)\n",
        "\n",
        "token2idx = {w:i for i, w in enumerate(idx2token)}\n",
        "\n",
        "glove.unk_init = lambda _: torch.zeros(glove.dim)\n",
        "embedding_vectors = torch.stack([glove[w] for w in idx2token])\n",
        "\n",
        "embedding = nn.Embedding(num_embeddings=embedding_vectors.size(0), \n",
        "                         embedding_dim=embedding_vectors.size(1),\n",
        "                         padding_idx=token2idx['<PAD>']).from_pretrained(embedding_vectors)\n",
        "embedding"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Embedding(75864, 100)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tYSFS-VyX9Zg",
        "outputId": "c2c213e6-42ea-4cfb-bfe6-5d659a8205f6"
      },
      "source": [
        "class Encode(object):\n",
        "  def __init__(self, token2idx):\n",
        "    self._token2idx = token2idx\n",
        "    self._vocab = set(token2idx.keys())\n",
        "\n",
        "  def __call__(self, data):\n",
        "    tokens, label = data['text'], data['label']\n",
        "    tokens = [token if token in embedding_vocab else '<UNK>' for token in tokens]\n",
        "    encoded = [token2idx[token] for token in tokens]\n",
        "    return {'text': encoded,\n",
        "            'label': label}\n",
        "\n",
        "class ToTensor(object):\n",
        "    def __call__(self, data):\n",
        "        text, label = data['text'], data['label']\n",
        "        return {'text': torch.LongTensor(text),\n",
        "                'label': torch.LongTensor([label])}\n",
        "\n",
        "from torchvision import transforms\n",
        "\n",
        "training_data = PolarityDataset(positive_dir=\"aclImdb/train/pos/\", \n",
        "                                negative_dir=\"aclImdb/train/neg/\",\n",
        "                                transform=transforms.Compose([Tokenize(), \n",
        "                                                              Encode(token2idx),\n",
        "                                                              ToTensor()]))\n",
        "print(training_data[0])"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'text': tensor([31312, 33833, 63004, 31129, 24588, 54649, 41888, 49607, 60197, 54537,\n",
            "          325, 16102, 20099, 11813, 65817, 32911, 68647, 18662, 54596, 48546,\n",
            "        54596,  2367, 54596, 53157, 54596, 52553, 61108, 60197, 55562, 16102,\n",
            "        72120, 19552, 54799, 32085, 73900, 47416, 59264, 65031, 55562, 11017,\n",
            "        31129, 40370, 60197, 32469, 49421, 24839, 38916, 14552, 24839,     0,\n",
            "        71421, 54799, 32085, 60406, 49607, 21062, 55562, 61622, 66257, 60545,\n",
            "        54596, 13924, 61079, 24839, 15005, 69813, 62964, 61622, 62192, 26942,\n",
            "        19552, 12285, 62964, 49421, 57511, 24839, 15815, 61622, 40245, 25394,\n",
            "        19552, 16102, 60854, 15156, 16102, 75388, 61622, 28209, 31908, 19552,\n",
            "        54799, 30519, 40206, 16102, 72120, 10968, 16032, 18087, 14360, 36750,\n",
            "        51620, 20487, 38916,     0, 16934,     0, 16934,     0, 24590, 49421,\n",
            "        48720,  9207,  7753, 54596, 50598, 10795, 50409, 55529, 46658, 54596,\n",
            "        19552, 25513,  6827, 50409, 21536, 69342, 27014, 33833, 50409, 14153,\n",
            "        10795, 54596, 36724, 15005, 21536, 21130, 11017, 16102, 28209, 24839,\n",
            "        35666, 47264, 63283, 54596, 15398, 54799, 75180, 32085, 23637, 21062,\n",
            "        39811, 50409, 55529, 33802, 35452, 29921, 39501,     0, 16934,     0,\n",
            "        16934,     0, 38893, 30899, 37201, 55562, 19552, 55562, 19552, 55562,\n",
            "         2250, 28106, 11993, 11017, 61622, 72120, 54596, 61622, 28158,  7583,\n",
            "        54596, 61622, 51755, 55666, 61108, 60197, 32911, 54799,  7702, 32085,\n",
            "        23637, 18087,   968, 51620, 35514,   325, 46710, 49421, 38893, 30899,\n",
            "        33804, 33061, 24839, 12998, 33061,  2250, 61622, 62327, 72120, 42564,\n",
            "        38916, 12212, 54596, 24839, 38893, 44780,   968, 51620, 74047, 70666,\n",
            "        19552, 70666, 30864,  8808, 53282, 51620, 18971, 19552, 21807, 15427,\n",
            "         9596, 29921, 11606, 40295, 60197]), 'label': tensor([1])}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "V0GbmgWIDLoT",
        "outputId": "6a337865-28fd-415c-f224-850be37719f7"
      },
      "source": [
        "def collate_with_padding(batch):\n",
        "  ret = {'label': torch.cat([sample['label'] for sample in batch])}\n",
        "  max_length = np.max([sample['text'].size(0) for sample in batch])\n",
        "  vectors = list()\n",
        "  for sample in batch:\n",
        "    if sample['text'].size(0) < max_length:\n",
        "      vectors.append(torch.cat([sample['text'], torch.ones(max_length-sample['text'].size(0))*token2idx['<PAD>']]))\n",
        "    else:\n",
        "      vectors.append(sample['text'])\n",
        "  ret['text'] = torch.stack(vectors).long()\n",
        "  return ret\n",
        "\n",
        "dataloader = torch.utils.data.DataLoader(training_data, batch_size=64, shuffle=True, num_workers=0, collate_fn=collate_with_padding, pin_memory=True)\n",
        "d = list(dataloader)[0]\n",
        "d"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'label': tensor([1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1,\n",
              "         0, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 0,\n",
              "         1, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 1]),\n",
              " 'text': tensor([[18690, 39970, 54596,  ...,     1,     1,     1],\n",
              "         [38893, 56408,  1196,  ...,     1,     1,     1],\n",
              "         [73677, 32085, 23637,  ...,     1,     1,     1],\n",
              "         ...,\n",
              "         [ 9263, 54596, 66093,  ...,     1,     1,     1],\n",
              "         [12998, 51354,    66,  ...,     1,     1,     1],\n",
              "         [12998, 51354, 12292,  ...,     1,     1,     1]])}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-mhJGmAbAen-"
      },
      "source": [
        "class ReviewPolarity(nn.Module):\n",
        "  def __init__(self, embedding, n_epochs=20, learning_rate=1, verbose=False):\n",
        "    \"\"\"\n",
        "    \"\"\"\n",
        "    super(ReviewPolarity, self).__init__()\n",
        "    self.n_epochs_ = n_epochs\n",
        "    self.verbose = verbose\n",
        "    # Define layers\n",
        "    self._embedding = embedding\n",
        "    self._conv3 = nn.Sequential(nn.Conv1d(self._embedding.embedding_dim, 100, kernel_size=3, padding=0),\n",
        "                                nn.ReLU(),\n",
        "                                nn.AdaptiveMaxPool1d(1))\n",
        "    self._conv4 = nn.Sequential(nn.Conv1d(self._embedding.embedding_dim, 100, kernel_size=4, padding=0),\n",
        "                                nn.ReLU(),\n",
        "                                nn.AdaptiveMaxPool1d(1))\n",
        "    self._conv5 = nn.Sequential(nn.Conv1d(self._embedding.embedding_dim, 100, kernel_size=5, padding=0),\n",
        "                                nn.ReLU(),\n",
        "                                nn.AdaptiveMaxPool1d(1))\n",
        "    self._dropout = nn.Dropout(0.5)\n",
        "    self._fc = nn.Sequential(nn.Linear(3*100, 2),\n",
        "                             nn.LogSoftmax(dim=1))\n",
        "    # Define optimiser and loss\n",
        "    self.learning_rate_ = learning_rate\n",
        "    self.optimizer_ = torch.optim.SGD(self.parameters(), lr=self.learning_rate_)\n",
        "    self.loss_function_ = nn.NLLLoss()\n",
        "    # For making nice plots\n",
        "    self.training_loss_ = list()\n",
        "    self.training_accuracy_ = list()\n",
        "    if torch.cuda.is_available():\n",
        "      self.cuda()\n",
        "\n",
        "\n",
        "  def forward(self, X):\n",
        "    # \"\"\"Forward pass, this is used for making predictions.\"\"\"\n",
        "    x = self._embedding(X)\n",
        "    x = x.transpose(1, 2)\n",
        "    c3 = self._conv3(x)\n",
        "    c4 = self._conv4(x)\n",
        "    c5 = self._conv5(x)\n",
        "    x = torch.cat([c3, c4, c5], dim=1)\n",
        "    x = x.view(x.size(0), -1)\n",
        "    x = self._dropout(x)\n",
        "    x = self._fc(x)\n",
        "    return x\n",
        "\n",
        "\n",
        "  def fit(self, X, y=None):\n",
        "    \"\"\"Train the model\"\"\"\n",
        "    # self._reset()\n",
        "    for epoch in range(self.n_epochs_):\n",
        "      if isinstance(X, torch.utils.data.dataloader.DataLoader):\n",
        "        for batch in tqdm(X, desc=\"Training epoch %i\" % epoch):\n",
        "          self._fit(batch['text'], batch['label'])\n",
        "      else:\n",
        "        self._fit(X, y)\n",
        "      if self.verbose:\n",
        "        if self.n_epochs_ < 50 or (epoch % 50) == 0:\n",
        "          print(\"Epoch %2i: Loss %.7f\" % (epoch, self._loss.item()))\n",
        "\n",
        "\n",
        "  def _fit(self, X, y):\n",
        "    \"\"\"A single step in the training\"\"\"\n",
        "    if torch.cuda.is_available():\n",
        "      X = X.cuda()\n",
        "      y = y.cuda()\n",
        "    self.zero_grad()              # Reset gradients\n",
        "    forward_pass_output = self(X) # Forward pass\n",
        "    self._loss = self.loss_function_(forward_pass_output, y) # Calculate loss\n",
        "    self._loss.backward()         # Backward pass\n",
        "    self.optimizer_.step()        # Run the optimizer one step\n",
        "    # Store values for user feedback\n",
        "    self.training_loss_.append(self._loss.item())\n",
        "    self.training_accuracy_.append(self.score(X, y))\n",
        "\n",
        "  def predict(self, X):\n",
        "    \"\"\"Predict labels\"\"\"\n",
        "    return np.asarray(np.argmax(self.predict_proba(X), axis=1), dtype=np.int)\n",
        "\n",
        "\n",
        "  def predict_proba(self, X):\n",
        "    \"\"\"Predict label probabilities\"\"\"\n",
        "    if torch.cuda.is_available():\n",
        "      X = X.cuda()\n",
        "    with torch.no_grad():         # Disable gradient calculations\n",
        "      model.eval()                # Disable dropout\n",
        "      x = self(X)                 # Forward pass without dropout\n",
        "      model.train()               # Enable dropout again, we need it during training\n",
        "    if torch.cuda.is_available():\n",
        "      x = x.cpu()\n",
        "    return np.exp(x.detach().numpy())\n",
        "\n",
        "\n",
        "  def score(self, X, y):\n",
        "    \"\"\"Get the accuracy of the model given the data\"\"\"\n",
        "    if torch.cuda.is_available():\n",
        "      y = y.cpu()\n",
        "    return np.sum(self.predict(X)==y.numpy())/len(y)\n",
        "\n",
        "model = ReviewPolarity(embedding=embedding, n_epochs=3, verbose=True, learning_rate=.1)"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9kJQDjjnm0n9",
        "outputId": "3101a9aa-6fe7-41a6-d99e-5b71dd1d1eda"
      },
      "source": [
        "model.fit(dataloader)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training epoch 0: 100%|██████████| 391/391 [00:19<00:00, 20.53it/s]\n",
            "Training epoch 1:   1%|          | 3/391 [00:00<00:18, 21.47it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch  0: Loss 0.5826154\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Training epoch 1: 100%|██████████| 391/391 [00:19<00:00, 20.27it/s]\n",
            "Training epoch 2:   1%|          | 3/391 [00:00<00:18, 20.91it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch  1: Loss 0.4585692\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Training epoch 2: 100%|██████████| 391/391 [00:18<00:00, 20.60it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch  2: Loss 0.4771730\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        },
        "id": "VASRthlGs33o",
        "outputId": "dc7ed80a-04c7-412f-b85f-50df0483cb01"
      },
      "source": [
        "batches = int(np.ceil(len(training_data)/dataloader.batch_size))\n",
        "\n",
        "fig = plt.figure(figsize=(16, 4))\n",
        "ax = fig.subplots(1, 2)\n",
        "ax[0].plot(model.training_loss_)\n",
        "ax[0].set_ylabel(\"Loss\")\n",
        "ax[0].set_xlabel(\"Batch\")\n",
        "ax[1].plot(100*np.asarray(model.training_accuracy_))\n",
        "ax[1].set_ylabel(\"Accuracy\")\n",
        "ax[1].set_xlabel(\"Batch\")\n",
        "fig.show()"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DtL3Rt71gGFo",
        "outputId": "fbff9d55-a488-461a-ca47-843ce4b04e9e"
      },
      "source": [
        "for p in model.parameters():\n",
        "  print(p.shape)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([75864, 100])\n",
            "torch.Size([100, 100, 3])\n",
            "torch.Size([100])\n",
            "torch.Size([100, 100, 4])\n",
            "torch.Size([100])\n",
            "torch.Size([100, 100, 5])\n",
            "torch.Size([100])\n",
            "torch.Size([2, 300])\n",
            "torch.Size([2])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Wh9ixkTPiAhv",
        "outputId": "f490b467-34b0-46d8-a9ba-1cff0c9282d7"
      },
      "source": [
        "test_data = PolarityDataset(positive_dir=\"aclImdb/test/pos/\", \n",
        "                            negative_dir=\"aclImdb/test/neg/\",\n",
        "                            transform=transforms.Compose([Tokenize(), \n",
        "                                                          Encode(token2idx),\n",
        "                                                          ToTensor()]))\n",
        "test_data_loader = torch.utils.data.DataLoader(test_data, batch_size=64, num_workers=0, \n",
        "                                               collate_fn=collate_with_padding)\n",
        "accuracy = np.average([model.score(batch['text'], batch['label']) for batch in tqdm(test_data_loader, desc=\"Calculating test set accuracy\")])\n",
        "print(\"Test set accuracy is %.1f%%\" % (100*accuracy))"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Calculating test set accuracy: 100%|██████████| 391/391 [00:12<00:00, 32.56it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Test set accuracy is 75.2%\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stderr"
        }
      ]
    }
  ]
}