ex2.ipynb

{"nbformat":4,"nbformat_minor":0,"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":"3.7.7"},"colab":{"name":"ex2.ipynb","provenance":[],"collapsed_sections":[]},"accelerator":"GPU"},"cells":[{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"1DAW5JWzCJ7Y","executionInfo":{"status":"ok","timestamp":1616732486045,"user_tz":-420,"elapsed":70551,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"985b71bf-2251-4ec1-8f66-677027dfdaf4"},"source":["from google.colab import drive\n","drive.mount('/content/drive')\n","%cd /content/drive/MyDrive/Question_Pairs_BiLSTM"],"execution_count":1,"outputs":[{"output_type":"stream","text":["Mounted at /content/drive\n","/content/drive/MyDrive/Question_Pairs_BiLSTM\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"BIpa0MkPB5a1","executionInfo":{"status":"ok","timestamp":1616732487740,"user_tz":-420,"elapsed":781,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["import pandas as pd\n","import numpy as np"],"execution_count":2,"outputs":[]},{"cell_type":"code","metadata":{"id":"V8WSUFOKoUWB","executionInfo":{"status":"ok","timestamp":1616732491702,"user_tz":-420,"elapsed":3382,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["data_train = pd.read_csv('data/train_dataset.csv')\n","data_val = pd.read_csv('data/val_dataset.csv')"],"execution_count":3,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"yfk0fsi0B5bF"},"source":["### BertWordPiece Tokenizer"]},{"cell_type":"code","metadata":{"id":"zvDD4Mj-do5T","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1616732501077,"user_tz":-420,"elapsed":7658,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"b3bb19c3-0d9c-42d7-caa8-1541ea235a4d"},"source":["!pip install -q transformers"],"execution_count":4,"outputs":[{"output_type":"stream","text":["\u001b[K     |████████████████████████████████| 2.0MB 7.5MB/s \n","\u001b[K     |████████████████████████████████| 3.2MB 56.0MB/s \n","\u001b[K     |████████████████████████████████| 890kB 54.8MB/s \n","\u001b[?25h  Building wheel for sacremoses (setup.py) ... \u001b[?25l\u001b[?25hdone\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"vdItZCoKB5bF","executionInfo":{"status":"ok","timestamp":1616732502251,"user_tz":-420,"elapsed":524,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["from tokenizers import BertWordPieceTokenizer\n","import os"],"execution_count":5,"outputs":[]},{"cell_type":"code","metadata":{"id":"kHTLaKtWDQNl","executionInfo":{"status":"ok","timestamp":1616732504652,"user_tz":-420,"elapsed":1757,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["f = open('data/questions.raw', 'w', encoding='utf-8')\n","for i in data_train.question1.to_list() + data_train.question2.to_list():\n","    f.write(i)\n","    f.write('\\n')\n","f.close()"],"execution_count":6,"outputs":[]},{"cell_type":"code","metadata":{"id":"WEpYfJMiB5bG","executionInfo":{"status":"ok","timestamp":1616732505963,"user_tz":-420,"elapsed":658,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["tokenizer = BertWordPieceTokenizer(lowercase=True)"],"execution_count":7,"outputs":[]},{"cell_type":"code","metadata":{"id":"6eAo8ZA5B5bH","executionInfo":{"status":"ok","timestamp":1616732524950,"user_tz":-420,"elapsed":18472,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["tokenizer.train('data/questions.raw')"],"execution_count":8,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"Zlztd8eutTyG","executionInfo":{"status":"ok","timestamp":1616732528828,"user_tz":-420,"elapsed":1009,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"e8dde27b-8e39-4ed7-bd78-c71d843184a7"},"source":["tokenizer.save_model('data/')"],"execution_count":9,"outputs":[{"output_type":"execute_result","data":{"text/plain":["['data/vocab.txt']"]},"metadata":{"tags":[]},"execution_count":9}]},{"cell_type":"code","metadata":{"id":"Niz80TuXwVgP","executionInfo":{"status":"ok","timestamp":1616732530109,"user_tz":-420,"elapsed":760,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["tokenizer = BertWordPieceTokenizer('data/vocab.txt')"],"execution_count":10,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"FbBsSrqxB5bI","executionInfo":{"status":"ok","timestamp":1616732531319,"user_tz":-420,"elapsed":532,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"bb62d308-30ee-47de-f23e-4bb2b5f6a5f7"},"source":["# test\n","print('tokens: {}'.format(tokenizer.encode(\"I'am Hungry!!!\").tokens))\n","print('ids: {}'.format(tokenizer.encode(\"I'am Hungry!!!\").ids))"],"execution_count":11,"outputs":[{"output_type":"stream","text":["tokens: ['[CLS]', 'i', \"'\", 'am', 'hungry', '!', '!', '!', '[SEP]']\n","ids: [2, 51, 11, 1561, 9183, 5, 5, 5, 3]\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"XwNFBoXMB5bI"},"source":["### Data Preprocessing"]},{"cell_type":"code","metadata":{"id":"G56ibQO-B5bJ","executionInfo":{"status":"ok","timestamp":1616732539224,"user_tz":-420,"elapsed":4169,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["from torch.utils import data\n","import torch\n","from torch.nn.utils.rnn import pad_sequence"],"execution_count":12,"outputs":[]},{"cell_type":"code","metadata":{"id":"eOgywQGwB5bJ","executionInfo":{"status":"ok","timestamp":1616732540373,"user_tz":-420,"elapsed":515,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["class CustomDataset(data.Dataset):\n","    def __init__(self, docs_1, docs_2, labels, tokenizer):\n","        self.docs_1 = docs_1\n","        self.docs_2 = docs_2\n","        self.labels = labels\n","        self.tokenizer = tokenizer\n","\n","    def __len__(self):\n","        return len(self.docs_1)\n","\n","    def __getitem__(self, item):\n","        docs_1 = str(self.docs_1[item])\n","        docs_2 = str(self.docs_2[item])\n","        labels = int(self.labels[item])\n","        encoded_docs_1 = self.tokenizer.encode(docs_1)\n","        encoded_docs_2 = self.tokenizer.encode(docs_2)\n","        return dict(\n","            input_ids_1 = torch.tensor(encoded_docs_1.ids),\n","            input_ids_2 = torch.tensor(encoded_docs_2.ids),\n","            labels = labels\n","            )"],"execution_count":13,"outputs":[]},{"cell_type":"code","metadata":{"id":"pLA4BxHKB5bK","executionInfo":{"status":"ok","timestamp":1616732541527,"user_tz":-420,"elapsed":551,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["train_data = CustomDataset(\n","    data_train.question1.to_numpy(),\n","    data_train.question2.to_numpy(), \n","    data_train.is_duplicate.to_numpy(),\n","    tokenizer\n","    )\n","validation_data = CustomDataset(\n","    data_val.question1.to_numpy(),\n","    data_val.question2.to_numpy(), \n","    data_val.is_duplicate.to_numpy(),\n","    tokenizer\n","    )"],"execution_count":14,"outputs":[]},{"cell_type":"code","metadata":{"id":"IWRWxWyKB5bK","executionInfo":{"status":"ok","timestamp":1616732542740,"user_tz":-420,"elapsed":671,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["def padding(batch):\n","    q1 = [q1['input_ids_1'] for q1 in batch]\n","    q2 = [q2['input_ids_2'] for q2 in batch]\n","    t = [t['labels'] for t in batch]\n","    len_q1 = [len(q1['input_ids_1']) for q1 in batch]\n","    len_q2 = [len(q2['input_ids_2']) for q2 in batch]\n","    q1_pad = pad_sequence(q1, batch_first=True, padding_value=0)\n","    q2_pad = pad_sequence(q2, batch_first=True, padding_value=0)\n","    return q1_pad, q2_pad, torch.tensor(t), torch.tensor(len_q1), torch.tensor(len_q2)"],"execution_count":15,"outputs":[]},{"cell_type":"code","metadata":{"id":"-Xcgb1FiB5bL","executionInfo":{"status":"ok","timestamp":1616732544063,"user_tz":-420,"elapsed":766,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["train_loader = data.DataLoader(train_data, batch_size = 64, shuffle=True, collate_fn=padding)\n","validation_loader = data.DataLoader(validation_data, batch_size = 128, collate_fn=padding)"],"execution_count":16,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"kQhBp_4mB5bL","executionInfo":{"status":"ok","timestamp":1616732545805,"user_tz":-420,"elapsed":996,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"5240fb7d-b58e-4bbc-85f2-6279ae8186ea"},"source":["for q1_pad, q2_pad, t, len_q1, len_q2 in train_loader:\n","    print('questions_1: {}, shape: {}'.format(q1_pad, q1_pad.shape))\n","    print('')\n","    print('len_questions_1: {}, shape: {}'.format(len_q1, len_q1.shape))\n","    print('')\n","    print('questions_2: {}, shape: {}'.format(q2_pad, q2_pad.shape))\n","    print('')\n","    print('len_questions_2: {}, shape: {}'.format(len_q2, len_q2.shape))\n","    print('')\n","    print('lables: {}, shape: {}'.format(t, t.shape))\n","    break"],"execution_count":17,"outputs":[{"output_type":"stream","text":["questions_1: tensor([[   2, 1396, 2686,  ...,    0,    0,    0],\n","        [   2, 1391, 1396,  ...,    0,    0,    0],\n","        [   2, 2018, 2597,  ...,    0,    0,    0],\n","        ...,\n","        [   2, 1546, 1623,  ...,    0,    0,    0],\n","        [   2, 1435, 1396,  ...,    0,    0,    0],\n","        [   2, 1391, 1396,  ...,    0,    0,    0]]), shape: torch.Size([64, 40])\n","\n","len_questions_1: tensor([18,  8, 21, 16, 16, 12, 40, 12,  9, 12, 17, 12, 17, 15, 16, 12,  8, 24,\n","        25, 10, 22,  8,  8, 13, 12, 22, 17, 14, 21, 14, 19, 13, 21,  8, 28, 16,\n","        31, 24, 23, 19, 10,  7, 16, 12, 16, 12, 17, 12, 18, 11,  8, 13, 17, 14,\n","        11, 13, 13, 20,  8, 12, 22,  7,  9, 11]), shape: torch.Size([64])\n","\n","questions_2: tensor([[    2,  1396, 21353,  ...,     0,     0,     0],\n","        [    2,  1411, 24654,  ...,     0,     0,     0],\n","        [    2,  1391,  2756,  ...,     0,     0,     0],\n","        ...,\n","        [    2,  1546,    11,  ...,     0,     0,     0],\n","        [    2,  8503,  1879,  ...,     0,     0,     0],\n","        [    2,  1391,  1396,  ...,     0,     0,     0]]), shape: torch.Size([64, 74])\n","\n","len_questions_2: tensor([14, 10, 10,  9, 24, 11, 21, 13,  9, 12, 20, 11, 14, 11, 14, 13,  9, 13,\n","        20, 17, 28,  8, 20, 16, 21, 21, 13,  9, 18, 14, 15, 28, 21, 11, 31, 15,\n","        29, 74, 19, 19, 14,  8, 15, 12, 15, 12, 17, 11,  7,  7, 12, 12, 12, 11,\n","        12, 15, 12, 14, 11, 12, 14, 12,  9, 16]), shape: torch.Size([64])\n","\n","lables: tensor([1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0,\n","        0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1,\n","        0, 0, 0, 1, 1, 0, 1, 1, 1, 1, 0, 1, 1, 0, 0, 0]), shape: torch.Size([64])\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"MFGA45-IB5bL","executionInfo":{"status":"ok","timestamp":1616732547364,"user_tz":-420,"elapsed":982,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"ece0ab14-42a0-40f4-91ce-9a46f9a9fcc2"},"source":["for q1_pad, q2_pad, t, len_q1, len_q2 in validation_loader:\n","    print('questions_1: {}, shape: {}'.format(q1_pad, q1_pad.shape))\n","    print('')\n","    print('len_questions_1: {}, shape: {}'.format(len_q1, len_q1.shape))\n","    print('')\n","    print('questions_2: {}, shape: {}'.format(q2_pad, q2_pad.shape))\n","    print('')\n","    print('len_questions_2: {}, shape: {}'.format(len_q2, len_q2.shape))\n","    print('')\n","    print('lables: {}, shape: {}'.format(t, t.shape))\n","    break"],"execution_count":18,"outputs":[{"output_type":"stream","text":["questions_1: tensor([[    2,  1403,  1398,  ...,     0,     0,     0],\n","        [    2,  1541,    43,  ...,     0,     0,     0],\n","        [    2,  1396, 19829,  ...,     0,     0,     0],\n","        ...,\n","        [    2,  1486,  1754,  ...,     0,     0,     0],\n","        [    2,  1391,  1396,  ...,     0,     0,     0],\n","        [    2,  1391,    11,  ...,     0,     0,     0]]), shape: torch.Size([128, 46])\n","\n","len_questions_1: tensor([11, 10, 11, 16, 10, 24, 22, 11, 11, 13, 20, 31, 16,  9, 31, 10, 21, 46,\n","        12, 11, 15, 11, 26,  9,  9, 10, 18, 11, 14, 14, 28, 12, 14, 11,  9, 10,\n","        15, 19, 16, 12, 18, 11, 11, 14, 12, 25, 16, 14, 13, 10, 15, 15, 17, 11,\n","        17, 11, 12, 10, 28, 14, 13,  9, 15, 16, 13, 20, 14, 11, 10, 10, 19, 16,\n","        11, 20, 14, 18,  7, 12,  7, 20, 12, 12,  9, 11, 17, 18, 22, 11, 12, 17,\n","        19, 22, 11, 18, 20,  9, 11, 13, 12, 12, 13, 17, 11, 11, 30,  8, 11, 11,\n","        10, 14, 15, 16, 12, 12, 11, 22, 12, 11, 23, 10, 10, 13, 22, 34, 13, 15,\n","        19, 18]), shape: torch.Size([128])\n","\n","questions_2: tensor([[    2,  1403,  1398,  ...,     0,     0,     0],\n","        [    2,  1403,  1417,  ...,     0,     0,     0],\n","        [    2,  1603, 19829,  ...,     0,     0,     0],\n","        ...,\n","        [    2,  2581, 14323,  ...,     0,     0,     0],\n","        [    2,  1391,  1570,  ...,     0,     0,     0],\n","        [    2,  1391,  1411,  ...,     0,     0,     0]]), shape: torch.Size([128, 46])\n","\n","len_questions_2: tensor([11, 14, 27, 27, 14, 19, 11, 14, 28, 13,  7, 39, 28,  9, 18, 11,  7, 28,\n","        13,  9, 14, 20, 14,  9,  8, 11, 13, 12, 12, 10, 46, 12, 10,  9, 11, 12,\n","        12, 18, 12, 18, 21, 23,  8, 12, 11, 32, 11, 14, 11, 10, 11, 17, 10, 13,\n","        41, 13, 14, 13, 26, 17, 10,  8, 10, 30, 16, 11, 18, 19, 11, 12, 13, 11,\n","        12, 18, 12, 28, 10, 17, 12, 21, 13, 18, 14, 11, 11, 16, 18, 40,  8, 17,\n","        15, 33, 10, 12, 20, 15, 14, 11, 12, 19, 13, 31, 14, 12, 27,  9,  8, 11,\n","        10, 17, 13, 15, 15, 11, 12,  9, 12, 18, 20, 13, 14, 13, 11, 12, 17, 16,\n","        16, 13]), shape: torch.Size([128])\n","\n","lables: tensor([0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1,\n","        1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0,\n","        1, 1, 0, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1, 0, 0,\n","        1, 0, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1,\n","        0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0,\n","        0, 0, 0, 0, 0, 0, 0, 0]), shape: torch.Size([128])\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"hz3fuPFWB5bM"},"source":["### Bidirectional LSTM"]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"bqz-MTEHB5bM","executionInfo":{"status":"ok","timestamp":1616732549368,"user_tz":-420,"elapsed":843,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"84f49a82-542e-40a6-f0bd-c7d5f4855d9e"},"source":["import torch.nn as nn\n","device = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\n","!nvidia-smi"],"execution_count":19,"outputs":[{"output_type":"stream","text":["Fri Mar 26 04:22:33 2021       \n","+-----------------------------------------------------------------------------+\n","| NVIDIA-SMI 460.56       Driver Version: 460.32.03    CUDA Version: 11.2     |\n","|-------------------------------+----------------------+----------------------+\n","| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n","| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n","|                               |                      |               MIG M. |\n","|===============================+======================+======================|\n","|   0  Tesla T4            Off  | 00000000:00:04.0 Off |                    0 |\n","| N/A   40C    P8     9W /  70W |      3MiB / 15109MiB |      0%      Default |\n","|                               |                      |                  N/A |\n","+-------------------------------+----------------------+----------------------+\n","                                                                               \n","+-----------------------------------------------------------------------------+\n","| Processes:                                                                  |\n","|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n","|        ID   ID                                                   Usage      |\n","|=============================================================================|\n","|  No running processes found                                                 |\n","+-----------------------------------------------------------------------------+\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"kvraNadRB5bM","executionInfo":{"status":"ok","timestamp":1616732551777,"user_tz":-420,"elapsed":802,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["class BiLSTM(nn.Module):\n","  def __init__(self, n_vocabs, embed_dims, n_lstm_units, n_lstm_layers, n_output_classes):\n","    super(BiLSTM, self).__init__()\n","    self.v = n_vocabs\n","    self.e = embed_dims\n","    self.u = n_lstm_units\n","    self.l = n_lstm_layers\n","    self.o = n_output_classes\n","    self.padd_idx = tokenizer.get_vocab()['[PAD]']\n","    self.embed = nn.Embedding(\n","        self.v,\n","        self.e,\n","        self.padd_idx\n","        )\n","    self.bilstm = nn.LSTM(\n","        self.e,\n","        self.u,\n","        self.l,\n","        batch_first = True,\n","        bidirectional = True,\n","        dropout = 0.5\n","        )\n","    self.linear = nn.Linear(\n","        self.u * 4,\n","        self.o\n","        )  \n","  \n","  def forward(self, X):\n","    # initial_hidden\n","    h0 = torch.zeros(self.l * 2, X[0].size(0), self.u).to(device)\n","    c0 = torch.zeros(self.l * 2, X[0].size(0), self.u).to(device)\n","    \n","    # embedding\n","    out1 = self.embed(X[0].to(device))\n","    out2 = self.embed(X[1].to(device))\n","\n","    # pack_padded_sequence\n","    out1 = nn.utils.rnn.pack_padded_sequence(out1, X[3].cpu(), batch_first=True, enforce_sorted=False)\n","    out2 = nn.utils.rnn.pack_padded_sequence(out2, X[4].cpu(), batch_first=True, enforce_sorted=False)\n","    \n","    # NxTxh, lxNxh\n","    out1, _ = self.bilstm(out1, (h0, c0))\n","    out2, _ = self.bilstm(out2, (h0, c0))\n","    \n","    # pad_packed_sequence\n","    out1, _ = nn.utils.rnn.pad_packed_sequence(out1, batch_first=True)\n","    out2, _ = nn.utils.rnn.pad_packed_sequence(out2, batch_first=True)\n","\n","    # take only the final time step\n","    out1 = out1[range(out1.shape[0]), X[3] - 1, :]\n","    out2 = out2[range(out2.shape[0]), X[4] - 1, :]\n","    \n","    # concatenate out1&2\n","    out = torch.cat((out1, out2), 1)\n","    \n","    # linear layer\n","    out = self.linear(out)\n","\n","    iout = torch.max(out, 1)[1]\n","    return iout, out"],"execution_count":20,"outputs":[]},{"cell_type":"code","metadata":{"id":"T2j4t2E_cfQQ","executionInfo":{"status":"ok","timestamp":1616732585918,"user_tz":-420,"elapsed":33135,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["train_labels = torch.cat([i[2] for i in train_loader])\n","negative = (train_labels == 0.).sum(dim=0).item()\n","positive = (train_labels == 1.).sum(dim=0).item()"],"execution_count":21,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"bzNEqS_wr6Vl","executionInfo":{"status":"ok","timestamp":1616732590189,"user_tz":-420,"elapsed":753,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"c38397ff-2168-42e6-b3e0-eb0fbd025614"},"source":["print(f'number data in training set {len(train_data)}')\n","print(f'number of negative data in training set {negative}')\n","print(f'number of positive data in training set {positive}')"],"execution_count":22,"outputs":[{"output_type":"stream","text":["number data in training set 323429\n","number of negative data in training set 203998\n","number of positive data in training set 119431\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"id":"Rwc0sJsffEPz"},"source":["class_weight :\n","\n","n = 203998 * (119431 / (119431 + 203998)) = 75329.3153613312\n","\n","p = 119431 * (203998 / (119431 + 203998)) = 75329.3153613312"]},{"cell_type":"code","metadata":{"id":"66js3dybbCF3","executionInfo":{"status":"ok","timestamp":1616732603154,"user_tz":-420,"elapsed":11564,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}}},"source":["torch.manual_seed(42)\n","model = BiLSTM(tokenizer.get_vocab_size(), 512, 512, 2, 2).to(device)\n","class_weight = torch.tensor([positive/(positive+negative), negative/(positive+negative)]).to(device)\n","criterion = nn.CrossEntropyLoss(weight=class_weight, reduction='mean').to(device)\n","optimizer = torch.optim.Adam(model.parameters(), lr = 1e-3)"],"execution_count":23,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"G5dVk_Okt7yR","executionInfo":{"status":"ok","timestamp":1616732612011,"user_tz":-420,"elapsed":773,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"c082927b-be41-458d-9fa5-30f64a6e23be"},"source":["print(model)"],"execution_count":24,"outputs":[{"output_type":"stream","text":["BiLSTM(\n","  (embed): Embedding(30000, 512, padding_idx=0)\n","  (bilstm): LSTM(512, 512, num_layers=2, batch_first=True, dropout=0.5, bidirectional=True)\n","  (linear): Linear(in_features=2048, out_features=2, bias=True)\n",")\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"6i_pGvc-1-4o","executionInfo":{"status":"ok","timestamp":1616735526017,"user_tz":-420,"elapsed":2912480,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"aba63d44-7a9c-4c1d-ad8d-0bfbf392923a"},"source":["import sys\n","from tqdm import tqdm\n","from datetime import datetime\n","\n","num_epochs = 5\n","losses = []\n","accuracies  = []\n","val_losses = []\n","val_accuracies = []\n","val_loss_min = np.inf\n","\n","for epoch in range(num_epochs):\n","    print('------------------------------------------------------------------------------------------')\n","    print('epoch: {}/{}:'.format(epoch + 1, num_epochs))   \n","    print('------------------------------------------------------------------------------------------')\n","    t0 = datetime.now()\n","\n","    train_tqdm_bar = tqdm(enumerate(train_loader), total = (len(train_loader)), leave = False, position = 0, file = sys.stdout, dynamic_ncols = True)\n","    val_tqdm_bar = tqdm(enumerate(validation_loader), total = (len(validation_loader)),  leave = False, position = 0, file = sys.stdout, dynamic_ncols = True)\n","\n","    running_loss = 0.0\n","    running_corrects = 0.0\n","    val_running_loss = 0.0\n","    val_running_corrects = 0.0\n","\n","    model.train()\n","    for idx, X in train_tqdm_bar:\n","        labels = X[2].to(device)\n","        iout, out = model(X)\n","        loss = criterion(out, labels)\n","        optimizer.zero_grad()\n","        loss.backward()\n","        torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm = 1)\n","        optimizer.step()\n","        running_loss += loss\n","        running_corrects += torch.sum(iout == labels)\n","        train_tqdm_bar.set_description(desc = 'train'.format(epoch + 1, num_epochs))\n","        batch_idx = (idx + 1) * 64\n","\n","        train_tqdm_bar.set_postfix(\n","            loss = running_loss.item() / batch_idx if idx + 1 < len(train_loader) else running_loss.item() / len(train_loader.dataset)\n","            ,acc = running_corrects.item() / batch_idx if idx + 1 < len(train_loader) else running_corrects.item() / len(train_loader.dataset)\n","            )\n","    \n","    model.eval()\n","    with torch.no_grad():\n","        for val_idx, v_X in val_tqdm_bar:\n","            v_labels = v_X[2].to(device)\n","            v_iout, v_out = model(v_X)\n","            v_loss = criterion(v_out, v_labels)\n","            val_running_loss += v_loss\n","            val_running_corrects += torch.sum(v_iout == v_labels)\n","            val_tqdm_bar.set_description('validate'.format(epoch + 1, num_epochs))\n","            val_batch_idx = (val_idx + 1) * 128\n","\n","            val_tqdm_bar.set_postfix(\n","                val_loss = val_running_loss.item() / val_batch_idx if val_idx + 1 < len(validation_loader) else val_running_loss.item() / len(validation_loader.dataset)\n","                ,val_acc = val_running_corrects.item() / val_batch_idx if val_idx + 1 < len(validation_loader) else val_running_corrects.item() / len(validation_loader.dataset)\n","                )\n","    \n","    epoch_loss = running_loss/len(train_loader.dataset)\n","    losses.append(epoch_loss)\n","    epoch_accuracy = running_corrects/len(train_loader.dataset)\n","    accuracies.append(epoch_accuracy)\n","    val_epoch_loss = val_running_loss/len(validation_loader.dataset)\n","    val_losses.append(val_epoch_loss)\n","    val_epoch_accuracy = val_running_corrects/len(validation_loader.dataset)\n","    val_accuracies.append(val_epoch_accuracy)\n","\n","    checkpoint = {\n","            'epoch': epoch + 1\n","            ,'state_dict': model.state_dict()\n","            ,'optimizer' : optimizer.state_dict()\n","            ,'val_loss_min' : val_epoch_loss\n","            }\n","    \n","    print('training loss: {:.4f}, acc: {:.2f}'.format(epoch_loss, epoch_accuracy))\n","    print('validation loss: {:.4f}, acc: {:.2f}'.format(val_epoch_loss, val_epoch_accuracy))\n","    print('elapsed time: {}'.format(str(datetime.now() - t0).split('.')[0]))\n","\n","    if val_epoch_loss <= val_loss_min:\n","      print('validation loss decreased from {:.4f} to {:.4f}, saving model...'.format(val_loss_min, val_epoch_loss))\n","      torch.save(checkpoint, 'checkpoint/ex2/question_pairs_lowest_val_loss_epoch_{}.pth'.format(epoch + 1))\n","      val_loss_min = val_epoch_loss"],"execution_count":25,"outputs":[{"output_type":"stream","text":["------------------------------------------------------------------------------------------\n","epoch: 1/5:\n","------------------------------------------------------------------------------------------\n","training loss: 0.0083, acc: 0.73\n","validation loss: 0.0039, acc: 0.77\n","elapsed time: 0:09:41\n","validation loss decreased from inf to 0.0039, saving model...\n","------------------------------------------------------------------------------------------\n","epoch: 2/5:\n","------------------------------------------------------------------------------------------\n","training loss: 0.0070, acc: 0.79\n","validation loss: 0.0038, acc: 0.78\n","elapsed time: 0:09:43\n","validation loss decreased from 0.0039 to 0.0038, saving model...\n","------------------------------------------------------------------------------------------\n","epoch: 3/5:\n","------------------------------------------------------------------------------------------\n","training loss: 0.0058, acc: 0.83\n","validation loss: 0.0039, acc: 0.78\n","elapsed time: 0:09:42\n","------------------------------------------------------------------------------------------\n","epoch: 4/5:\n","------------------------------------------------------------------------------------------\n","training loss: 0.0047, acc: 0.86\n","validation loss: 0.0042, acc: 0.77\n","elapsed time: 0:09:40\n","------------------------------------------------------------------------------------------\n","epoch: 5/5:\n","------------------------------------------------------------------------------------------\n","training loss: 0.0039, acc: 0.89\n","validation loss: 0.0049, acc: 0.78\n","elapsed time: 0:09:41\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"FGH5nSZCFZPR","colab":{"base_uri":"https://localhost:8080/","height":277},"executionInfo":{"status":"ok","timestamp":1616735526798,"user_tz":-420,"elapsed":2909956,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"6382deff-b895-4edc-e66a-3015c3df3301"},"source":["import matplotlib.pyplot as plt\n","plt.style.use('seaborn-darkgrid')\n","plt.figure()\n","plt.plot(accuracies, color = 'magenta')\n","plt.plot(val_accuracies, color = '#606060')\n","plt.title('Model Accuracy')\n","plt.legend(['train_acc', 'val_acc'], loc = 'lower right')\n","plt.grid(axis = 'y', c = 'black', alpha = 0.2)\n","plt.grid(axis = 'x', c = 'black', alpha = 0.2)"],"execution_count":26,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 432x288 with 1 Axes>"]},"metadata":{"tags":[]}}]},{"cell_type":"code","metadata":{"id":"BLlgEBleGHxw","colab":{"base_uri":"https://localhost:8080/","height":277},"executionInfo":{"status":"ok","timestamp":1616735526800,"user_tz":-420,"elapsed":2909152,"user":{"displayName":"rafi stars","photoUrl":"","userId":"08632319744334099596"}},"outputId":"f41daf10-1ff6-4716-de1f-e800d3057bbf"},"source":["plt.figure()\n","plt.plot(losses, color = 'magenta')\n","plt.plot(val_losses, color = '#606060')\n","plt.title('Model Loss')\n","plt.legend(['train_loss', 'val_loss'], loc = 'upper right')\n","plt.grid(axis = 'y', c = 'black', alpha = 0.2)\n","plt.grid(axis = 'x', c = 'black', alpha = 0.2)"],"execution_count":27,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 432x288 with 1 Axes>"]},"metadata":{"tags":[]}}]}]}