blstm_dfw_snli.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "F2hv12elHUkk"
   },
   "source": [
    "# Training BiLSTM on SNLI using DFW"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "executionInfo": {
     "elapsed": 2985,
     "status": "ok",
     "timestamp": 1619449382782,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "x4cJFGwS4Foe",
    "outputId": "68ef3bf9-6232-45bc-8eb5-00d3962c6b40"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mounted at /content/drive\n",
      "/content/drive/MyDrive/DL_project\n",
      "/content/drive/MyDrive/DL_project\n"
     ]
    }
   ],
   "source": [
    "from google.colab import drive\n",
    "drive.mount('/content/drive', force_remount=True)\n",
    "%cd /content/drive/MyDrive/DL_project/\n",
    "!pwd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "executionInfo": {
     "elapsed": 2231,
     "status": "ok",
     "timestamp": 1619449418895,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "nLURgE-65F61"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torchvision\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torchvision.datasets import CIFAR10\n",
    "from torchvision.datasets import CIFAR100\n",
    "from torch.utils.data.dataloader import DataLoader\n",
    "from torchvision.transforms import *\n",
    "import os\n",
    "import time\n",
    "import gc\n",
    "import shutil\n",
    "import h5py\n",
    "\n",
    "from dfw.dfw import DFW\n",
    "from dfw.dfw.losses import set_smoothing_enabled\n",
    "from dfw.dfw.losses import MultiClassHingeLoss\n",
    "from dfw.experiments.InferSent.data import get_nli, get_batch, build_vocab\n",
    "from dfw.experiments.InferSent.models import NLINet\n",
    "\n",
    "from snli_train_utils import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "executionInfo": {
     "elapsed": 169116,
     "status": "ok",
     "timestamp": 1619449588226,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "g9ywtAAt-8qH",
    "outputId": "d4abc3a1-96ea-442f-de35-e08473e04578"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "** TRAIN DATA : Found 549367 pairs of train sentences.\n",
      "** DEV DATA : Found 9842 pairs of dev sentences.\n",
      "** TEST DATA : Found 9824 pairs of test sentences.\n",
      "Found 38957(/43479) words with glove vectors\n",
      "Vocab size : 38957\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:12: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
      "  if sys.path[0] == '':\n"
     ]
    }
   ],
   "source": [
    "GLOVE_PATH = \"dfw/experiments/InferSent/dataset/GloVe/glove.840B.300d.txt\"\n",
    "nlipath = 'dfw/experiments/InferSent/dataset/SNLI'\n",
    "train, valid, test = get_nli(nlipath)\n",
    "word_vec = build_vocab(train['s1'] + train['s2'] +\n",
    "                       valid['s1'] + valid['s2'] +\n",
    "                       test['s1'] + test['s2'], GLOVE_PATH)\n",
    "\n",
    "for split in ['s1', 's2']:\n",
    "    for data_type in ['train', 'valid', 'test']:\n",
    "        eval(data_type)[split] = np.array([['<s>'] +\n",
    "            [word for word in sent.split() if word in word_vec] +\n",
    "            ['</s>'] for sent in eval(data_type)[split]])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "executionInfo": {
     "elapsed": 166768,
     "status": "ok",
     "timestamp": 1619449588228,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "EPhF30-jBOfu"
   },
   "outputs": [],
   "source": [
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "word_emb_dim = 300\n",
    "config_nli_model = {\n",
    "    'n_words'        :  len(word_vec)         ,\n",
    "    'word_emb_dim'   :  word_emb_dim   ,\n",
    "    'enc_lstm_dim'   :  2048   ,\n",
    "    'n_enc_layers'   :  1   ,\n",
    "    'dpout_model'    :  0.    ,\n",
    "    'dpout_fc'       :  0.       ,\n",
    "    'fc_dim'         :  512         ,\n",
    "    'bsize'          :  64     ,\n",
    "    'n_classes'      :  3      ,\n",
    "    'pool_type'      :  'max'      ,\n",
    "    'nonlinear_fc'   :  0   ,\n",
    "    'encoder_type'   :  'BLSTMEncoder'   ,\n",
    "    'use_cuda'       :  True                  ,\n",
    "\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "cB6cyM6FNnbO",
    "outputId": "5cd91f1a-ab4b-4c2f-c97a-0c31e2a2b7e3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda Tesla T4\n",
      "Epoch:0   Loss:0.339   Val loss:1.354   Accuracy:45.094   Best val loss:1.354   Time taken:324.687\n",
      "Epoch:1   Loss:0.188   Val loss:1.248   Accuracy:47.156   Best val loss:1.248   Time taken:333.782\n",
      "Epoch:2   Loss:0.151   Val loss:1.438   Accuracy:43.844   Best val loss:1.248   Time taken:331.068\n",
      "Epoch:3   Loss:0.168   Val loss:1.326   Accuracy:43.609   Best val loss:1.248   Time taken:328.06\n",
      "Epoch:4   Loss:0.156   Val loss:1.275   Accuracy:50.938   Best val loss:1.248   Time taken:331.373\n",
      "Epoch:5   Loss:0.14   Val loss:1.258   Accuracy:49.766   Best val loss:1.248   Time taken:332.239\n",
      "Epoch:6   Loss:0.146   Val loss:1.438   Accuracy:48.547   Best val loss:1.248   Time taken:334.514\n",
      "Epoch:7   Loss:0.132   Val loss:1.333   Accuracy:51.484   Best val loss:1.248   Time taken:335.417\n",
      "Epoch:8   Loss:0.127   Val loss:1.175   Accuracy:53.688   Best val loss:1.175   Time taken:330.917\n",
      "Epoch:9   Loss:0.125   Val loss:1.204   Accuracy:53.719   Best val loss:1.175   Time taken:330.517\n",
      "Epoch:10   Loss:0.12   Val loss:1.208   Accuracy:57.141   Best val loss:1.175   Time taken:330.337\n",
      "Epoch:11   Loss:0.12   Val loss:1.243   Accuracy:54.594   Best val loss:1.175   Time taken:331.767\n",
      "Epoch:12   Loss:0.127   Val loss:1.02   Accuracy:63.75   Best val loss:1.02   Time taken:331.548\n",
      "Epoch:13   Loss:0.126   Val loss:1.193   Accuracy:56.734   Best val loss:1.02   Time taken:328.635\n",
      "Epoch:14   Loss:0.123   Val loss:1.164   Accuracy:56.25   Best val loss:1.02   Time taken:332.658\n",
      "Epoch:15   Loss:0.109   Val loss:1.179   Accuracy:54.969   Best val loss:1.02   Time taken:330.456\n",
      "Epoch:16   Loss:0.116   Val loss:1.153   Accuracy:53.062   Best val loss:1.02   Time taken:330.898\n",
      "Epoch:17   Loss:0.119   Val loss:1.234   Accuracy:49.141   Best val loss:1.02   Time taken:328.213\n",
      "Epoch:18   Loss:0.111   Val loss:1.316   Accuracy:51.859   Best val loss:1.02   Time taken:334.558\n",
      "Epoch:19   Loss:0.112   Val loss:1.337   Accuracy:51.938   Best val loss:1.02   Time taken:328.683\n",
      "Epoch:20   Loss:0.105   Val loss:1.151   Accuracy:55.938   Best val loss:1.02   Time taken:332.668\n",
      "Epoch:21   Loss:0.113   Val loss:1.101   Accuracy:61.266   Best val loss:1.02   Time taken:336.196\n",
      "Epoch:22   Loss:0.119   Val loss:1.181   Accuracy:51.203   Best val loss:1.02   Time taken:334.865\n",
      "Epoch:23   Loss:0.11   Val loss:1.242   Accuracy:57.469   Best val loss:1.02   Time taken:328.281\n",
      "Epoch:24   Loss:0.106   Val loss:1.175   Accuracy:52.031   Best val loss:1.02   Time taken:325.342\n",
      "Epoch:25   Loss:0.105   Val loss:1.178   Accuracy:53.844   Best val loss:1.02   Time taken:326.559\n",
      "Epoch:26   Loss:0.106   Val loss:1.133   Accuracy:56.672   Best val loss:1.02   Time taken:333.004\n",
      "Epoch:27   Loss:0.107   Val loss:1.335   Accuracy:51.031   Best val loss:1.02   Time taken:326.994\n",
      "Epoch:28   Loss:0.099   Val loss:1.342   Accuracy:53.938   Best val loss:1.02   Time taken:331.132\n",
      "Epoch:29   Loss:0.096   Val loss:1.139   Accuracy:53.609   Best val loss:1.02   Time taken:330.551\n"
     ]
    }
   ],
   "source": [
    "trainer = Net_Trainer(device, 'nli_dfw_snli', config_nli_model, word_vec, train, valid, test, lr=1.0, optm='dfw', loss='svm', batch_size=64, epochs=100, \n",
    "                          lr_decay=0.99, lr_shrink=5.0, weight_decay=0, mom=0, smooth=False)\n",
    "best_epoch, losses, errors, accs, testacc = trainer.train()\n",
    "plot_train(losses, errors, accs)\n",
    "print('Best epoch:',best_epoch)\n",
    "print('Test accuracy:',testacc,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 816
    },
    "executionInfo": {
     "elapsed": 8213168,
     "status": "error",
     "timestamp": 1619439527030,
     "user": {
      "displayName": "Mansi Khamkar",
      "photoUrl": "",
      "userId": "07087059808651280926"
     },
     "user_tz": -330
    },
    "id": "nuge5xfw4-iU",
    "outputId": "ebb8b3e5-519d-4110-aadd-b11051570b33"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda Tesla T4\n",
      "Epoch:30   Loss:0.103   Val loss:1.117   Accuracy:58.922   Best val loss:1.02   Time taken:340.386\n",
      "Epoch:31   Loss:0.096   Val loss:1.217   Accuracy:58.297   Best val loss:1.02   Time taken:341.433\n",
      "Epoch:32   Loss:0.089   Val loss:1.103   Accuracy:56.703   Best val loss:1.02   Time taken:340.224\n",
      "Epoch:33   Loss:0.11   Val loss:1.184   Accuracy:54.734   Best val loss:1.02   Time taken:344.373\n",
      "Epoch:34   Loss:0.1   Val loss:1.199   Accuracy:51.234   Best val loss:1.02   Time taken:340.976\n",
      "Epoch:35   Loss:0.095   Val loss:1.182   Accuracy:52.859   Best val loss:1.02   Time taken:344.155\n",
      "Epoch:36   Loss:0.102   Val loss:1.186   Accuracy:59.0   Best val loss:1.02   Time taken:339.629\n",
      "Epoch:37   Loss:0.101   Val loss:1.174   Accuracy:51.453   Best val loss:1.02   Time taken:341.703\n",
      "Epoch:38   Loss:0.091   Val loss:1.218   Accuracy:56.984   Best val loss:1.02   Time taken:339.084\n",
      "Epoch:39   Loss:0.102   Val loss:1.185   Accuracy:56.094   Best val loss:1.02   Time taken:343.576\n",
      "Epoch:40   Loss:0.088   Val loss:1.406   Accuracy:50.312   Best val loss:1.02   Time taken:336.719\n",
      "Epoch:41   Loss:0.088   Val loss:1.299   Accuracy:54.234   Best val loss:1.02   Time taken:338.441\n",
      "Epoch:42   Loss:0.073   Val loss:1.456   Accuracy:53.406   Best val loss:1.02   Time taken:339.394\n",
      "Epoch:43   Loss:0.094   Val loss:1.373   Accuracy:54.953   Best val loss:1.02   Time taken:337.9\n",
      "Epoch:44   Loss:0.085   Val loss:1.283   Accuracy:55.297   Best val loss:1.02   Time taken:337.763\n",
      "Epoch:45   Loss:0.091   Val loss:1.309   Accuracy:54.0   Best val loss:1.02   Time taken:339.944\n",
      "Epoch:46   Loss:0.087   Val loss:1.278   Accuracy:56.531   Best val loss:1.02   Time taken:339.369\n",
      "Epoch:47   Loss:0.098   Val loss:1.289   Accuracy:52.188   Best val loss:1.02   Time taken:333.956\n",
      "Epoch:48   Loss:0.09   Val loss:1.172   Accuracy:56.016   Best val loss:1.02   Time taken:336.905\n",
      "Epoch:49   Loss:0.094   Val loss:1.165   Accuracy:57.922   Best val loss:1.02   Time taken:333.903\n",
      "Epoch:50   Loss:0.103   Val loss:1.194   Accuracy:58.047   Best val loss:1.02   Time taken:333.411\n",
      "Epoch:51   Loss:0.088   Val loss:1.264   Accuracy:60.031   Best val loss:1.02   Time taken:334.933\n",
      "Epoch:52   Loss:0.092   Val loss:1.209   Accuracy:58.656   Best val loss:1.02   Time taken:334.603\n",
      "Epoch:53   Loss:0.099   Val loss:1.293   Accuracy:51.359   Best val loss:1.02   Time taken:338.169\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "ignored",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-6-e895fd2701bb>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      4\u001b[0m trainer = Net_Trainer(device, 'nli_dfw_snli', config_nli_model, word_vec, train, valid, test, load_ckpt='v1_ckpt_nli_dfw_snli', lr=1.0, optm='dfw', loss='svm', batch_size=64, epochs=100, \n\u001b[1;32m      5\u001b[0m                           lr_decay=0.99, lr_shrink=5.0, weight_decay=0, mom=0, smooth=False)\n\u001b[0;32m----> 6\u001b[0;31m \u001b[0mbest_epoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlosses\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0merrors\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maccs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtestacc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtrainer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      7\u001b[0m \u001b[0mplot_train\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlosses\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0merrors\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maccs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      8\u001b[0m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'Best epoch:'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mbest_epoch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/content/drive/My Drive/DL_project/snli_train_utils.py\u001b[0m in \u001b[0;36mtrain\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    155\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mepoch\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstart_epoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mepochs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    156\u001b[0m             \u001b[0mstart\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtime\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtime\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 157\u001b[0;31m             \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain_epoch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mepoch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    158\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mepoch\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m1\u001b[0m \u001b[0;34m!=\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    159\u001b[0m                 \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'Epoch:'\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mepoch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m'  Loss:'\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mround\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m3\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/content/drive/My Drive/DL_project/snli_train_utils.py\u001b[0m in \u001b[0;36mtrain_epoch\u001b[0;34m(self, cur_epoch)\u001b[0m\n\u001b[1;32m    218\u001b[0m             \u001b[0;31m# print('*',proc_items)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    219\u001b[0m             \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mempty_cache\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 220\u001b[0;31m             \u001b[0mgc\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcollect\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    221\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mlosses\u001b[0m\u001b[0;34m/\u001b[0m\u001b[0mbatches\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    222\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# Conitnue training by loading checkpoint\n",
    "# Best epoch = epoch with least val loss\n",
    "trainer = Net_Trainer(device, 'nli_dfw_snli', config_nli_model, word_vec, train, valid, test, load_ckpt='v1_ckpt_nli_dfw_snli', lr=1.0, optm='dfw', loss='svm', batch_size=64, epochs=100, \n",
    "                          lr_decay=0.99, lr_shrink=5.0, weight_decay=0, mom=0, smooth=False)\n",
    "best_epoch, losses, errors, accs, testacc = trainer.train()\n",
    "plot_train(losses, errors, accs)\n",
    "print('Best epoch:',best_epoch)\n",
    "print('Test accuracy:',testacc,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 624
    },
    "executionInfo": {
     "elapsed": 60292,
     "status": "ok",
     "timestamp": 1619447618152,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "Qj7eVPAHiY81",
    "outputId": "d2db8794-6b4f-4355-f80a-0c4eab33aa37"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda Tesla K80\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best epoch: 12\n",
      "Test accuracy: 57.859375 %\n"
     ]
    }
   ],
   "source": [
    "# This cell is used only to plot graphs till epoch 40\n",
    "trainer = Net_Trainer(device, 'nli_dfw_snli', config_nli_model, word_vec, train, valid, test, load_ckpt='v1_ckpt_nli_dfw_snli_ep39', lr=1.0, optm='dfw', loss='svm', \n",
    "                      batch_size=64, epochs=40, lr_decay=0.99, lr_shrink=5.0, weight_decay=0, mom=0, smooth=False)\n",
    "best_epoch, losses, errors, accs, testacc = trainer.train()\n",
    "plot_train(losses, errors, accs)\n",
    "print('Best epoch:',best_epoch)\n",
    "print('Test accuracy:',testacc,'%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "executionInfo": {
     "elapsed": 70303,
     "status": "ok",
     "timestamp": 1619449658552,
     "user": {
      "displayName": "MANSI KHAMKAR",
      "photoUrl": "https://lh3.googleusercontent.com/a-/AOh14GhMFZb_-ee8qqaB-dV_gK7g_R_at5uh-HS8mBMQ=s64",
      "userId": "07862734149699594801"
     },
     "user_tz": -330
    },
    "id": "WXnom5Zqq51A",
    "outputId": "a23b5bdb-3465-44f3-f58a-877ae0e6c9c3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test accuracy of best model: 58.171875\n"
     ]
    }
   ],
   "source": [
    "best_acc = test_model(device, config_nli_model, 'v1_ckpt_nli_dfw_snli_best', test, 64, word_vec)\n",
    "print('Test accuracy of best model:',best_acc)"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "blstm_dfw_snli.ipynb",
   "provenance": []
  },
  "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}