IMDB Parameter Sweeps.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Background Stuff\n",
    "Explain what we are doing here"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from glob import glob\n",
    "import os\n",
    "import sys\n",
    "from datetime import datetime\n",
    "import matplotlib.pyplot as plt\n",
    "from scipy.stats import describe\n",
    "import re\n",
    "from collections import Counter\n",
    "\n",
    "from src.utils import prep_data, train_and_evaluate_model, get_x_y\n",
    "\n",
    "SEED = 2018"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running trial 1 of 7\n",
      "Epoch 1/1\n",
      "5000/5000 [==============================] - 55s 11ms/step - loss: 0.6767 - acc: 0.5748\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 2 of 7\n",
      "Epoch 1/2\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6899 - acc: 0.5288\n",
      "Epoch 2/2\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6463 - acc: 0.6368\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 3 of 7\n",
      "Epoch 1/5\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6844 - acc: 0.5814\n",
      "Epoch 2/5\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.7108 - acc: 0.5356\n",
      "Epoch 3/5\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6824 - acc: 0.5626\n",
      "Epoch 4/5\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6717 - acc: 0.5892\n",
      "Epoch 5/5\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6734 - acc: 0.5830\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 4 of 7\n",
      "Epoch 1/10\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6742 - acc: 0.5760\n",
      "Epoch 2/10\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6025 - acc: 0.6792\n",
      "Epoch 3/10\n",
      "5000/5000 [==============================] - 50s 10ms/step - loss: 0.5907 - acc: 0.6886\n",
      "Epoch 4/10\n",
      "5000/5000 [==============================] - 49s 10ms/step - loss: 0.5616 - acc: 0.7136\n",
      "Epoch 5/10\n",
      "5000/5000 [==============================] - 51s 10ms/step - loss: 0.5697 - acc: 0.7052\n",
      "Epoch 6/10\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5578 - acc: 0.7126\n",
      "Epoch 7/10\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5443 - acc: 0.7296\n",
      "Epoch 8/10\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5381 - acc: 0.7332\n",
      "Epoch 9/10\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5996 - acc: 0.6738\n",
      "Epoch 10/10\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5735 - acc: 0.7034\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 5 of 7\n",
      "Epoch 1/15\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.7020 - acc: 0.5246\n",
      "Epoch 2/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6839 - acc: 0.5708\n",
      "Epoch 3/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6771 - acc: 0.5696\n",
      "Epoch 4/15\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6528 - acc: 0.6264\n",
      "Epoch 5/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6159 - acc: 0.6706\n",
      "Epoch 6/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6048 - acc: 0.6774\n",
      "Epoch 7/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5942 - acc: 0.6864\n",
      "Epoch 8/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5874 - acc: 0.6922\n",
      "Epoch 9/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5792 - acc: 0.7004\n",
      "Epoch 10/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5851 - acc: 0.6988\n",
      "Epoch 11/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5666 - acc: 0.7080\n",
      "Epoch 12/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5492 - acc: 0.7258\n",
      "Epoch 13/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5401 - acc: 0.7294\n",
      "Epoch 14/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5494 - acc: 0.7280\n",
      "Epoch 15/15\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5442 - acc: 0.7268\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 6 of 7\n",
      "Epoch 1/20\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6837 - acc: 0.5668\n",
      "Epoch 2/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6078 - acc: 0.6738\n",
      "Epoch 3/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5806 - acc: 0.6980\n",
      "Epoch 4/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5799 - acc: 0.6986\n",
      "Epoch 5/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5737 - acc: 0.7058\n",
      "Epoch 6/20\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6098 - acc: 0.6826\n",
      "Epoch 7/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5781 - acc: 0.6978\n",
      "Epoch 8/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5590 - acc: 0.7144\n",
      "Epoch 9/20\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5525 - acc: 0.7252\n",
      "Epoch 10/20\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5318 - acc: 0.7398\n",
      "Epoch 11/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5405 - acc: 0.7340\n",
      "Epoch 12/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5339 - acc: 0.7426\n",
      "Epoch 13/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5216 - acc: 0.7466\n",
      "Epoch 14/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5300 - acc: 0.7402\n",
      "Epoch 15/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5312 - acc: 0.7398\n",
      "Epoch 16/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.7151 - acc: 0.5682\n",
      "Epoch 17/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6792 - acc: 0.5770\n",
      "Epoch 18/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6622 - acc: 0.6090\n",
      "Epoch 19/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6481 - acc: 0.6208\n",
      "Epoch 20/20\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6453 - acc: 0.6266\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 7 of 7\n",
      "Epoch 1/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6779 - acc: 0.5686\n",
      "Epoch 2/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6309 - acc: 0.6616\n",
      "Epoch 3/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5925 - acc: 0.6944\n",
      "Epoch 4/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5744 - acc: 0.7050\n",
      "Epoch 5/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5745 - acc: 0.7024\n",
      "Epoch 6/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5741 - acc: 0.7062\n",
      "Epoch 7/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5553 - acc: 0.7182\n",
      "Epoch 8/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5485 - acc: 0.7320\n",
      "Epoch 9/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5429 - acc: 0.7274\n",
      "Epoch 10/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5420 - acc: 0.7258\n",
      "Epoch 11/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5477 - acc: 0.7286\n",
      "Epoch 12/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5408 - acc: 0.7338\n",
      "Epoch 13/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5345 - acc: 0.7368\n",
      "Epoch 14/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5287 - acc: 0.7402\n",
      "Epoch 15/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5224 - acc: 0.7486\n",
      "Epoch 16/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5385 - acc: 0.7332\n",
      "Epoch 17/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5449 - acc: 0.7268\n",
      "Epoch 18/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5243 - acc: 0.7442\n",
      "Epoch 19/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5212 - acc: 0.7424\n",
      "Epoch 20/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5097 - acc: 0.7510\n",
      "Epoch 21/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5037 - acc: 0.7572\n",
      "Epoch 22/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5011 - acc: 0.7570\n",
      "Epoch 23/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5035 - acc: 0.7566\n",
      "Epoch 24/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5082 - acc: 0.7490\n",
      "Epoch 25/30\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.4937 - acc: 0.7648\n",
      "Epoch 26/30\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5039 - acc: 0.7558\n",
      "Epoch 27/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.4970 - acc: 0.7608\n",
      "Epoch 28/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5014 - acc: 0.7594\n",
      "Epoch 29/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5415 - acc: 0.7244\n",
      "Epoch 30/30\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5071 - acc: 0.7540\n",
      "5000/5000 [==============================] - 12s 2ms/step\n"
     ]
    }
   ],
   "source": [
    "number_epochs = [1, 2, 5, 10, 15, 20, 30]\n",
    "num_trials = len(number_epochs)\n",
    "scores = []\n",
    "vocab_size = 200\n",
    "max_review_len = 300\n",
    "\n",
    "for counter, epochs in enumerate(number_epochs):\n",
    "        print('Running trial ' + str(counter + 1) + ' of ' + str(num_trials))\n",
    "        x_train_pad_sub, y_train_sub, x_test_pad_sub, y_test_sub = prep_data(\n",
    "            vocab_size=vocab_size,\n",
    "            max_review_len=max_review_len\n",
    "        )\n",
    "\n",
    "        scores.append(\n",
    "            train_and_evaluate_model(\n",
    "                x_train=x_train_pad_sub,\n",
    "                y_train=y_train_sub,\n",
    "                x_test=x_test_pad_sub,\n",
    "                y_test=y_test_sub,\n",
    "                num_epochs=epochs,\n",
    "                max_review_len=max_review_len,\n",
    "                verbose=1\n",
    "            )\n",
    "        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x640 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(9, 8), dpi= 80, facecolor='w', edgecolor='k')\n",
    "width = 0.25\n",
    "x_range = np.arange(len(scores))\n",
    "plt.bar(x_range, scores, width)\n",
    "plt.xticks(x_range, number_epochs)\n",
    "plt.ylim([0, 100])\n",
    "plt.xlabel('Number of Epochs')\n",
    "plt.ylabel('Accuracy %')\n",
    "plt.title('Number of Epochs vs Accuracy (5k Samples Used)')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Comparison of Max Review Lenghts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_review_lengths = [10, 50, 100, 200, 300, 400, 500]\n",
    "vocab_sizes = [100, 250, 500]\n",
    "scores = {}\n",
    "counter = 1\n",
    "num_trials = len(max_review_lengths) * len(vocab_sizes)\n",
    "\n",
    "for vocab_size in vocab_sizes:\n",
    "    scores[vocab_size] = []\n",
    "    for max_review in max_review_lengths:\n",
    "        print('Running trial ' + str(counter) + ' of ' + str(num_trials))\n",
    "        x_train_pad_sub, y_train_sub, x_test_pad_sub, y_test_sub = prep_data(\n",
    "            vocab_size=vocab_size,\n",
    "            max_review_len=max_review\n",
    "        )\n",
    "\n",
    "        scores[vocab_size].append(\n",
    "            train_and_evaluate_model(\n",
    "                x_train=x_train_pad_sub,\n",
    "                y_train=y_train_sub,\n",
    "                x_test=x_test_pad_sub,\n",
    "                y_test=y_test_sub,\n",
    "                num_epochs=2,\n",
    "                max_review_len=max_review,\n",
    "                verbose=0\n",
    "            )\n",
    "        )\n",
    "\n",
    "        counter += 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fig = plt.figure(figsize=(9, 8), dpi= 80, facecolor='w', edgecolor='k')\n",
    "x_range = np.arange(len(next(iter(scores.values()))))\n",
    "width = 0.25\n",
    "legend_values = []\n",
    "\n",
    "for counter, vocab_size in enumerate(scores):\n",
    "    plt.bar(x_range + width*(counter - 1), scores[vocab_size], width)\n",
    "    legend_values.append('Vocab Size ' + str(vocab_size))\n",
    "    \n",
    "plt.xticks(x_range, max_review_lengths)\n",
    "plt.ylim([0, 100])\n",
    "plt.xlabel('Max Review Lengths')\n",
    "plt.ylabel('Accuracy %')\n",
    "plt.title('Length of Reviews vs Accuracy (5k Samples Used)')\n",
    "plt.legend(legend_values)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_lengths = [8, 32, 64, 128, 256]\n",
    "scores = []\n",
    "vocab_size = 500\n",
    "max_review_len = 300\n",
    "\n",
    "x_train_pad_sub, y_train_sub, x_test_pad_sub, y_test_sub = prep_data(\n",
    "            vocab_size=vocab_size,\n",
    "            max_review_len=max_review\n",
    "        )\n",
    "\n",
    "for counter, embed_len in enumerate(embed_lengths):\n",
    "        print('Running trial ' + str(counter + 1) + ' of ' + str(len(embed_lengths)))\n",
    "        scores.append(\n",
    "            train_and_evaluate_model(\n",
    "                x_train=x_train_pad_sub,\n",
    "                y_train=y_train_sub,\n",
    "                x_test=x_test_pad_sub,\n",
    "                y_test=y_test_sub,\n",
    "                vocab_size=vocab_size,\n",
    "                embed_length=embed_len,\n",
    "                num_epochs=3,\n",
    "                max_review_len=max_review,\n",
    "                verbose=1\n",
    "            )\n",
    "        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fig = plt.figure(figsize=(9, 8), dpi= 80, facecolor='w', edgecolor='k')\n",
    "width = 0.25\n",
    "x_range = np.arange(len(scores))\n",
    "plt.bar(x_range, scores, width)\n",
    "plt.xticks(x_range, embed_lengths)\n",
    "plt.ylim([0, 100])\n",
    "plt.xlabel('Embedded Vector Length')\n",
    "plt.ylabel('Accuracy %')\n",
    "plt.title('Embedded Vector Length vs Accuracy (5k Samples Used)')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_reviews = [1000, 2500, 5000, 10000, 25000]\n",
    "scores = []\n",
    "vocab_size = 200\n",
    "max_review_len = 300\n",
    "\n",
    "for counter, n_samples in enumerate(num_reviews):\n",
    "        print('Running trial ' + str(counter + 1) + ' of ' + str(len(num_reviews)))\n",
    "        x_train_pad_sub, y_train_sub, x_test_pad_sub, y_test_sub = prep_data(\n",
    "            vocab_size=vocab_size,\n",
    "            max_review_len=max_review,\n",
    "            n_samples=n_samples\n",
    "        )\n",
    "\n",
    "        scores.append(\n",
    "            train_and_evaluate_model(\n",
    "                x_train=x_train_pad_sub,\n",
    "                y_train=y_train_sub,\n",
    "                x_test=x_test_pad_sub,\n",
    "                y_test=y_test_sub,\n",
    "                vocab_size=vocab_size,\n",
    "                num_epochs=3,\n",
    "                max_review_len=max_review,\n",
    "                verbose=1\n",
    "            )\n",
    "        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fig = plt.figure(figsize=(9, 8), dpi= 80, facecolor='w', edgecolor='k')\n",
    "width = 0.25\n",
    "x_range = np.arange(len(scores))\n",
    "plt.bar(x_range, scores, width)\n",
    "plt.xticks(x_range, num_reviews)\n",
    "plt.ylim([0, 100])\n",
    "plt.xlabel('Number of Samples')\n",
    "plt.ylabel('Accuracy %')\n",
    "plt.title('Number of Samples vs Accuracy')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running trial 1 of 3\n",
      "Epoch 1/3\n",
      "5000/5000 [==============================] - 54s 11ms/step - loss: 0.6784 - acc: 0.5754\n",
      "Epoch 2/3\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6115 - acc: 0.6708\n",
      "Epoch 3/3\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.5856 - acc: 0.6984\n",
      "5000/5000 [==============================] - 12s 2ms/step\n",
      "Running trial 2 of 3\n",
      "Epoch 1/3\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6929 - acc: 0.5468\n",
      "Epoch 2/3\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.6812 - acc: 0.5680\n",
      "Epoch 3/3\n",
      "5000/5000 [==============================] - 52s 10ms/step - loss: 0.6898 - acc: 0.5462\n",
      "5000/5000 [==============================] - 13s 3ms/step\n",
      "Running trial 3 of 3\n",
      "Epoch 1/3\n",
      "5000/5000 [==============================] - 58s 12ms/step - loss: 0.6940 - acc: 0.5432\n",
      "Epoch 2/3\n",
      "5000/5000 [==============================] - 56s 11ms/step - loss: 0.6162 - acc: 0.6646\n",
      "Epoch 3/3\n",
      "5000/5000 [==============================] - 53s 11ms/step - loss: 0.5806 - acc: 0.6954\n",
      "5000/5000 [==============================] - 12s 2ms/step\n"
     ]
    }
   ],
   "source": [
    "lstm_lengths = [10, 25, 50] #, 100, 150, 200, 500]\n",
    "scores = []\n",
    "vocab_size = 200\n",
    "max_review_len = 300\n",
    "\n",
    "for counter, lstm_len in enumerate(lstm_lengths):\n",
    "        print('Running trial ' + str(counter + 1) + ' of ' + str(len(lstm_lengths)))\n",
    "        x_train_pad_sub, y_train_sub, x_test_pad_sub, y_test_sub = prep_data(\n",
    "            vocab_size=vocab_size,\n",
    "            max_review_len=max_review_len,\n",
    "            n_samples=5000\n",
    "        )\n",
    "\n",
    "        scores.append(\n",
    "            train_and_evaluate_model(\n",
    "                x_train=x_train_pad_sub,\n",
    "                y_train=y_train_sub,\n",
    "                x_test=x_test_pad_sub,\n",
    "                y_test=y_test_sub,\n",
    "                num_epochs=3,\n",
    "                lstm_len=lstm_len,\n",
    "                max_review_len=max_review_len,\n",
    "                verbose=1\n",
    "            )\n",
    "        )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x640 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(9, 8), dpi= 80, facecolor='w', edgecolor='k')\n",
    "width = 0.25\n",
    "x_range = np.arange(len(scores))\n",
    "plt.bar(x_range, scores, width)\n",
    "plt.xticks(x_range, lstm_lengths)\n",
    "plt.ylim([0, 100])\n",
    "plt.xlabel('LSTM Length')\n",
    "plt.ylabel('Accuracy %')\n",
    "plt.title('LSTM Length vs Accuracy')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}