train_tune_test.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In a [another notebook](./explore_data_and_model.ipynb) we explored the Fashion MNIST image data set and the CNN model used to classify them. \n",
    "\n",
    "In this notebook, we'll train the model, tune it a bit, and finally test it.\n",
    "\n",
    "We're going to use Amazon's Sagemaker cloud service to overcome local resource limitations.\n",
    "We'll take advantage of its convenient [Python SDK](https://sagemaker.readthedocs.io/en/stable/) which manages AWS resources for us behind the scenes. A [`ml.p3.2xlarge` instance](https://aws.amazon.com/sagemaker/pricing/instance-types/) will significantly speed up training and choosing managed spot instances will yield considerable savings (usually 60-70%)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/anaconda3/envs/fashion/lib/python3.7/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import os\n",
    "import sagemaker\n",
    "import boto3\n",
    "import h5py\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sagemaker.tensorflow import TensorFlow\n",
    "from sagemaker.tuner import IntegerParameter, CategoricalParameter, ContinuousParameter, HyperparameterTuner\n",
    "from cnn import FashionMNISTCNN as fmc\n",
    "\n",
    "# filter out FutureWarnings\n",
    "from warnings import simplefilter\n",
    "simplefilter(action='ignore', category=FutureWarning)\n",
    "\n",
    "# Supress Tensorflow Warnings\n",
    "import tensorflow.compat.v1.logging as logging\n",
    "logging.set_verbosity(logging.ERROR)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "toc-hr-collapsed": true
   },
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Sagemaker will run the training script inside a (prebuilt) Docker container and will pull data from an s3 bucket we specify. The container will be torn down on completion of the training job but we can send container files to an s3 bucket before that. In particular, we'll send the validation accuracy improvement checkpoints and training history generated by our training script `train_script_sagemaker.py`.\n",
    "\n",
    "We'll use the same s3 bucket for all of this. First we'll upload local data to the bucket, then create a directory for storing keras checkpoints and history. Finally we'll specify a path for the \"model artifacts\" of the training job, i.e. anything saved in the `opt/ml/model` directory of the training job container. In our case, this is just the Tensorflow serving model.\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Set up s3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Session info\n",
    "sess = sagemaker.Session()\n",
    "role_name = '<YOUR IAM ROLE NAME>'\n",
    "bucket_name = '<YOUR BUCKET NAME>'\n",
    "\n",
    "# upload data to s3\n",
    "training_input_path   = sess.upload_data('data/train.hdf5', bucket=bucket_name, key_prefix='data')\n",
    "validation_input_path = sess.upload_data('data/val.hdf5', bucket=bucket_name, key_prefix='data')\n",
    "test_input_path = sess.upload_data('data/test.hdf5', bucket=bucket_name, key_prefix='data')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create checkpoint directory in s3\n",
    "try:\n",
    "    with open('models/keras_checkpoints/dummy.txt', 'x') as f:\n",
    "        f.write('This is a dummy file')\n",
    "except OSError:\n",
    "    pass\n",
    "\n",
    "checks_output_path = sess.upload_data('models/keras_checkpoints/dummy.txt', bucket=bucket_name, key_prefix='keras-checkpoints')\n",
    "checks_output_path = os.path.dirname(checks_output_path)\n",
    "\n",
    "# s3 path for job output\n",
    "job_output_path = 's3://{}/'.format(bucket_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Run a single training job"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll run a single Sagemaker training job using the [default model]('./explore_data_and_model.ipynb/#keras-cnn-model-for-classification')\n",
    "\n",
    "We use a `sagemaker.tensorflow.Tensorflow` estimator for this training job. We'll track loss and accuracy metrics for both training and validation data, which keras tracks by default. \n",
    "\n",
    "Note that our output path for keras checkpoints gets passed in as a hyperparameter."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# objective and metric\n",
    "metric_definitions = [{'Name': 'acc',\n",
    "                       'Regex': 'acc: ([0-9\\\\.]+)'},\n",
    "                      {'Name': 'val_acc',\n",
    "                       'Regex': 'val_acc: ([0-9\\\\.]+)'},\n",
    "                      {'Name': 'loss',\n",
    "                       'Regex': 'loss: ([0-9\\\\.]+)'},\n",
    "                      {'Name': 'val_loss',\n",
    "                       'Regex': 'val_loss: ([0-9\\\\.]+)'}]\n",
    "\n",
    "\n",
    "hyperparameters = {'epochs': 100, 'batch-size': 100, 'drop-rate': 0.5,\n",
    "                   'checks-out-path': checks_output_path}\n",
    "\n",
    "# create sagemaker estimator\n",
    "tf_estimator = TensorFlow(entry_point='train_script_sagemaker.py', \n",
    "                          role=role_name,\n",
    "                          train_volume_size=5,\n",
    "                          train_instance_count=1, \n",
    "                          train_instance_type='ml.p3.2xlarge',\n",
    "                          train_use_spot_instances=True,\n",
    "                          train_max_wait=86400,\n",
    "                          output_path=job_output_path,\n",
    "                          framework_version='1.14', \n",
    "                          py_version='py3',\n",
    "                          script_mode=True,\n",
    "                          hyperparameters=hyperparameters,\n",
    "                          metric_definitions=metric_definitions\n",
    "                         )\n",
    "\n",
    "paths = {'train': training_input_path, 'val': validation_input_path,\n",
    "         'test': test_input_path, 'checks': checks_output_path}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# train estimator asynchronously\n",
    "tf_estimator.fit(paths, wait=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "toc-hr-collapsed": true
   },
   "source": [
    "### Evaluate training job"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Download Keras checkpoints and history from s3"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we pull the keras checkpoints and history down from s3."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "def download_checks_from_s3(checks_output_path):\n",
    "    s3_resource = boto3.resource('s3')\n",
    "    bucket_name = os.path.dirname(checks_output_path).split('//')[1]\n",
    "    prefix = os.path.basename(checks_output_path)\n",
    "    bucket = s3_resource.Bucket(bucket_name)\n",
    "\n",
    "    for obj in bucket.objects.filter(Prefix = prefix):\n",
    "        local_dir = 'models/keras_checkpoints'\n",
    "        if not os.path.exists(local_dir):\n",
    "            os.makedirs(local_dir)\n",
    "        local_file = os.path.join(local_dir, \n",
    "                                  os.path.basename(obj.key))\n",
    "        bucket.download_file(obj.key, local_file)\n",
    "\n",
    "# delete any preexisting checkpoints\n",
    "! rm models/keras_checkpoints/*\n",
    "download_checks_from_s3(checks_output_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Analyze training history "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll plot the keras training history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val_loss</th>\n",
       "      <th>val_acc</th>\n",
       "      <th>loss</th>\n",
       "      <th>acc</th>\n",
       "      <th>lr</th>\n",
       "      <th>epoch</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.333004</td>\n",
       "      <td>0.8778</td>\n",
       "      <td>0.520144</td>\n",
       "      <td>0.82050</td>\n",
       "      <td>0.001</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.263404</td>\n",
       "      <td>0.9033</td>\n",
       "      <td>0.316812</td>\n",
       "      <td>0.88600</td>\n",
       "      <td>0.001</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.247057</td>\n",
       "      <td>0.9091</td>\n",
       "      <td>0.268965</td>\n",
       "      <td>0.90370</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.297089</td>\n",
       "      <td>0.8980</td>\n",
       "      <td>0.240364</td>\n",
       "      <td>0.91154</td>\n",
       "      <td>0.001</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.251827</td>\n",
       "      <td>0.9074</td>\n",
       "      <td>0.221876</td>\n",
       "      <td>0.92054</td>\n",
       "      <td>0.001</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   val_loss  val_acc      loss      acc     lr  epoch\n",
       "0  0.333004   0.8778  0.520144  0.82050  0.001      1\n",
       "1  0.263404   0.9033  0.316812  0.88600  0.001      2\n",
       "2  0.247057   0.9091  0.268965  0.90370  0.001      3\n",
       "3  0.297089   0.8980  0.240364  0.91154  0.001      4\n",
       "4  0.251827   0.9074  0.221876  0.92054  0.001      5"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history_df = pd.read_csv('models/keras_checkpoints/FashionMNISTCNN-history.csv')\n",
    "history_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def plot_history(history_df):\n",
    "    fig, ax = plt.subplots(1, 2, figsize=(15, 10))\n",
    "\n",
    "    plt.subplot(1, 2, 1)\n",
    "    plt.plot('epoch', 'loss', data=history_df, label='train_loss')\n",
    "    plt.plot('epoch', 'val_loss', data=history_df, label='val_loss')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('Loss')\n",
    "    plt.title('Training and validation loss')\n",
    "    plt.legend()\n",
    "    \n",
    "    plt.subplot(1, 2, 2)\n",
    "    plt.plot('epoch', 'acc', data=history_df, label='train_acc')\n",
    "    plt.plot('epoch', 'val_acc', data=history_df, label='val_acc')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('Accuracy')\n",
    "    plt.title('Training and validation accuracy')\n",
    "    plt.legend()\n",
    "    \n",
    "plot_history(history_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Maximum training accuracy epoch: \n",
      "val_loss     0.275882\n",
      "val_acc      0.935300\n",
      "loss         0.040384\n",
      "acc          0.985260\n",
      "lr           0.001000\n",
      "epoch       42.000000\n",
      "Name: 41, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "acc_max = history_df.loc[history_df['acc'].idxmax(), :]\n",
    "print('Maximum training accuracy epoch: \\n{}'.format(acc_max))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Maximum validation accuracy epoch: \n",
      "val_loss     0.237604\n",
      "val_acc      0.938600\n",
      "loss         0.055796\n",
      "acc          0.979620\n",
      "lr           0.001000\n",
      "epoch       32.000000\n",
      "Name: 31, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "val_acc_max = history_df.loc[history_df['val_acc'].idxmax(), :]\n",
    "print('Maximum validation accuracy epoch: \\n{}'.format(val_acc_max))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val_acc</th>\n",
       "      <th>acc</th>\n",
       "      <th>lr</th>\n",
       "      <th>epoch</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.9386</td>\n",
       "      <td>0.97962</td>\n",
       "      <td>0.001</td>\n",
       "      <td>32</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>0.9369</td>\n",
       "      <td>0.98394</td>\n",
       "      <td>0.001</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.9367</td>\n",
       "      <td>0.97210</td>\n",
       "      <td>0.001</td>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>0.9365</td>\n",
       "      <td>0.98308</td>\n",
       "      <td>0.001</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.9357</td>\n",
       "      <td>0.97814</td>\n",
       "      <td>0.001</td>\n",
       "      <td>30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>0.9356</td>\n",
       "      <td>0.98280</td>\n",
       "      <td>0.001</td>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>0.9353</td>\n",
       "      <td>0.98526</td>\n",
       "      <td>0.001</td>\n",
       "      <td>42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.9352</td>\n",
       "      <td>0.97368</td>\n",
       "      <td>0.001</td>\n",
       "      <td>25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>0.9351</td>\n",
       "      <td>0.98288</td>\n",
       "      <td>0.001</td>\n",
       "      <td>37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.9350</td>\n",
       "      <td>0.97824</td>\n",
       "      <td>0.001</td>\n",
       "      <td>31</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.9345</td>\n",
       "      <td>0.97972</td>\n",
       "      <td>0.001</td>\n",
       "      <td>34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.9342</td>\n",
       "      <td>0.98230</td>\n",
       "      <td>0.001</td>\n",
       "      <td>35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.9340</td>\n",
       "      <td>0.97988</td>\n",
       "      <td>0.001</td>\n",
       "      <td>33</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.9339</td>\n",
       "      <td>0.96342</td>\n",
       "      <td>0.001</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.9339</td>\n",
       "      <td>0.97800</td>\n",
       "      <td>0.001</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.9339</td>\n",
       "      <td>0.97544</td>\n",
       "      <td>0.001</td>\n",
       "      <td>28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.9335</td>\n",
       "      <td>0.96752</td>\n",
       "      <td>0.001</td>\n",
       "      <td>21</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.9334</td>\n",
       "      <td>0.97144</td>\n",
       "      <td>0.001</td>\n",
       "      <td>23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.9317</td>\n",
       "      <td>0.97508</td>\n",
       "      <td>0.001</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>0.9313</td>\n",
       "      <td>0.98254</td>\n",
       "      <td>0.001</td>\n",
       "      <td>38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.9310</td>\n",
       "      <td>0.95914</td>\n",
       "      <td>0.001</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.9306</td>\n",
       "      <td>0.95532</td>\n",
       "      <td>0.001</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.9306</td>\n",
       "      <td>0.96310</td>\n",
       "      <td>0.001</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.9300</td>\n",
       "      <td>0.96974</td>\n",
       "      <td>0.001</td>\n",
       "      <td>22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.9297</td>\n",
       "      <td>0.98172</td>\n",
       "      <td>0.001</td>\n",
       "      <td>36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.9288</td>\n",
       "      <td>0.96678</td>\n",
       "      <td>0.001</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.9281</td>\n",
       "      <td>0.94712</td>\n",
       "      <td>0.001</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.9281</td>\n",
       "      <td>0.93736</td>\n",
       "      <td>0.001</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.9279</td>\n",
       "      <td>0.97398</td>\n",
       "      <td>0.001</td>\n",
       "      <td>26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.9265</td>\n",
       "      <td>0.96062</td>\n",
       "      <td>0.001</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.9261</td>\n",
       "      <td>0.95488</td>\n",
       "      <td>0.001</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.9258</td>\n",
       "      <td>0.94466</td>\n",
       "      <td>0.001</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.9216</td>\n",
       "      <td>0.94084</td>\n",
       "      <td>0.001</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.9114</td>\n",
       "      <td>0.94970</td>\n",
       "      <td>0.001</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.9091</td>\n",
       "      <td>0.90370</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.9074</td>\n",
       "      <td>0.92054</td>\n",
       "      <td>0.001</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.9033</td>\n",
       "      <td>0.88600</td>\n",
       "      <td>0.001</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.9031</td>\n",
       "      <td>0.92906</td>\n",
       "      <td>0.001</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.9002</td>\n",
       "      <td>0.93370</td>\n",
       "      <td>0.001</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.8980</td>\n",
       "      <td>0.91154</td>\n",
       "      <td>0.001</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.8947</td>\n",
       "      <td>0.92276</td>\n",
       "      <td>0.001</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.8778</td>\n",
       "      <td>0.82050</td>\n",
       "      <td>0.001</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    val_acc      acc     lr  epoch\n",
       "31   0.9386  0.97962  0.001     32\n",
       "40   0.9369  0.98394  0.001     41\n",
       "23   0.9367  0.97210  0.001     24\n",
       "38   0.9365  0.98308  0.001     39\n",
       "29   0.9357  0.97814  0.001     30\n",
       "39   0.9356  0.98280  0.001     40\n",
       "41   0.9353  0.98526  0.001     42\n",
       "24   0.9352  0.97368  0.001     25\n",
       "36   0.9351  0.98288  0.001     37\n",
       "30   0.9350  0.97824  0.001     31\n",
       "33   0.9345  0.97972  0.001     34\n",
       "34   0.9342  0.98230  0.001     35\n",
       "32   0.9340  0.97988  0.001     33\n",
       "18   0.9339  0.96342  0.001     19\n",
       "28   0.9339  0.97800  0.001     29\n",
       "27   0.9339  0.97544  0.001     28\n",
       "20   0.9335  0.96752  0.001     21\n",
       "22   0.9334  0.97144  0.001     23\n",
       "26   0.9317  0.97508  0.001     27\n",
       "37   0.9313  0.98254  0.001     38\n",
       "15   0.9310  0.95914  0.001     16\n",
       "14   0.9306  0.95532  0.001     15\n",
       "17   0.9306  0.96310  0.001     18\n",
       "21   0.9300  0.96974  0.001     22\n",
       "35   0.9297  0.98172  0.001     36\n",
       "19   0.9288  0.96678  0.001     20\n",
       "11   0.9281  0.94712  0.001     12\n",
       "8    0.9281  0.93736  0.001      9\n",
       "25   0.9279  0.97398  0.001     26\n",
       "16   0.9265  0.96062  0.001     17\n",
       "13   0.9261  0.95488  0.001     14\n",
       "10   0.9258  0.94466  0.001     11\n",
       "9    0.9216  0.94084  0.001     10\n",
       "12   0.9114  0.94970  0.001     13\n",
       "2    0.9091  0.90370  0.001      3\n",
       "4    0.9074  0.92054  0.001      5\n",
       "1    0.9033  0.88600  0.001      2\n",
       "6    0.9031  0.92906  0.001      7\n",
       "7    0.9002  0.93370  0.001      8\n",
       "3    0.8980  0.91154  0.001      4\n",
       "5    0.8947  0.92276  0.001      6\n",
       "0    0.8778  0.82050  0.001      1"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Validation accuracy epochs in descending order\n",
    "history_df.drop(columns=['val_loss', 'loss']).sort_values(by='val_acc', ascending=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We note that $93\\%$ accuracy first occured roughly during epochs 15-18, and didn't improve much thereafter.\n",
    "\n",
    "The last epoch where improvement occured was epoch 32, and since the default model has an early stopping patience of 10 epochs, we know it didn't improve from epochs 32-42 and training stopped after epoch 42."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val_loss</th>\n",
       "      <th>loss</th>\n",
       "      <th>lr</th>\n",
       "      <th>epoch</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.200692</td>\n",
       "      <td>0.122466</td>\n",
       "      <td>0.001</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.204977</td>\n",
       "      <td>0.168892</td>\n",
       "      <td>0.001</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.206213</td>\n",
       "      <td>0.151945</td>\n",
       "      <td>0.001</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.206824</td>\n",
       "      <td>0.144087</td>\n",
       "      <td>0.001</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.207615</td>\n",
       "      <td>0.112607</td>\n",
       "      <td>0.001</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.212421</td>\n",
       "      <td>0.088360</td>\n",
       "      <td>0.001</td>\n",
       "      <td>21</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.214470</td>\n",
       "      <td>0.099285</td>\n",
       "      <td>0.001</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.216433</td>\n",
       "      <td>0.124686</td>\n",
       "      <td>0.001</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.220042</td>\n",
       "      <td>0.107955</td>\n",
       "      <td>0.001</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.221858</td>\n",
       "      <td>0.100688</td>\n",
       "      <td>0.001</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.223961</td>\n",
       "      <td>0.078725</td>\n",
       "      <td>0.001</td>\n",
       "      <td>23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.223996</td>\n",
       "      <td>0.162515</td>\n",
       "      <td>0.001</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.224511</td>\n",
       "      <td>0.073269</td>\n",
       "      <td>0.001</td>\n",
       "      <td>25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.229332</td>\n",
       "      <td>0.091966</td>\n",
       "      <td>0.001</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.231225</td>\n",
       "      <td>0.082998</td>\n",
       "      <td>0.001</td>\n",
       "      <td>22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.231908</td>\n",
       "      <td>0.076089</td>\n",
       "      <td>0.001</td>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.236227</td>\n",
       "      <td>0.065881</td>\n",
       "      <td>0.001</td>\n",
       "      <td>28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.237604</td>\n",
       "      <td>0.055796</td>\n",
       "      <td>0.001</td>\n",
       "      <td>32</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.241979</td>\n",
       "      <td>0.068059</td>\n",
       "      <td>0.001</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.243946</td>\n",
       "      <td>0.061642</td>\n",
       "      <td>0.001</td>\n",
       "      <td>30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.244725</td>\n",
       "      <td>0.055597</td>\n",
       "      <td>0.001</td>\n",
       "      <td>33</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.247057</td>\n",
       "      <td>0.268965</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.251827</td>\n",
       "      <td>0.221876</td>\n",
       "      <td>0.001</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.252233</td>\n",
       "      <td>0.061005</td>\n",
       "      <td>0.001</td>\n",
       "      <td>31</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.255708</td>\n",
       "      <td>0.055532</td>\n",
       "      <td>0.001</td>\n",
       "      <td>34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.255919</td>\n",
       "      <td>0.135069</td>\n",
       "      <td>0.001</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>0.256263</td>\n",
       "      <td>0.045249</td>\n",
       "      <td>0.001</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.256528</td>\n",
       "      <td>0.070083</td>\n",
       "      <td>0.001</td>\n",
       "      <td>26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>0.258491</td>\n",
       "      <td>0.046385</td>\n",
       "      <td>0.001</td>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>0.258976</td>\n",
       "      <td>0.047897</td>\n",
       "      <td>0.001</td>\n",
       "      <td>37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.259441</td>\n",
       "      <td>0.048968</td>\n",
       "      <td>0.001</td>\n",
       "      <td>35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>0.261749</td>\n",
       "      <td>0.047258</td>\n",
       "      <td>0.001</td>\n",
       "      <td>38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.262236</td>\n",
       "      <td>0.060982</td>\n",
       "      <td>0.001</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.263404</td>\n",
       "      <td>0.316812</td>\n",
       "      <td>0.001</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>0.264150</td>\n",
       "      <td>0.045217</td>\n",
       "      <td>0.001</td>\n",
       "      <td>39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.272420</td>\n",
       "      <td>0.194994</td>\n",
       "      <td>0.001</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>0.275882</td>\n",
       "      <td>0.040384</td>\n",
       "      <td>0.001</td>\n",
       "      <td>42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.279350</td>\n",
       "      <td>0.050247</td>\n",
       "      <td>0.001</td>\n",
       "      <td>36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.285503</td>\n",
       "      <td>0.182704</td>\n",
       "      <td>0.001</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.295849</td>\n",
       "      <td>0.209872</td>\n",
       "      <td>0.001</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.297089</td>\n",
       "      <td>0.240364</td>\n",
       "      <td>0.001</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.333004</td>\n",
       "      <td>0.520144</td>\n",
       "      <td>0.001</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    val_loss      loss     lr  epoch\n",
       "14  0.200692  0.122466  0.001     15\n",
       "8   0.204977  0.168892  0.001      9\n",
       "10  0.206213  0.151945  0.001     11\n",
       "11  0.206824  0.144087  0.001     12\n",
       "15  0.207615  0.112607  0.001     16\n",
       "20  0.212421  0.088360  0.001     21\n",
       "18  0.214470  0.099285  0.001     19\n",
       "13  0.216433  0.124686  0.001     14\n",
       "16  0.220042  0.107955  0.001     17\n",
       "17  0.221858  0.100688  0.001     18\n",
       "22  0.223961  0.078725  0.001     23\n",
       "9   0.223996  0.162515  0.001     10\n",
       "24  0.224511  0.073269  0.001     25\n",
       "19  0.229332  0.091966  0.001     20\n",
       "21  0.231225  0.082998  0.001     22\n",
       "23  0.231908  0.076089  0.001     24\n",
       "27  0.236227  0.065881  0.001     28\n",
       "31  0.237604  0.055796  0.001     32\n",
       "26  0.241979  0.068059  0.001     27\n",
       "29  0.243946  0.061642  0.001     30\n",
       "32  0.244725  0.055597  0.001     33\n",
       "2   0.247057  0.268965  0.001      3\n",
       "4   0.251827  0.221876  0.001      5\n",
       "30  0.252233  0.061005  0.001     31\n",
       "33  0.255708  0.055532  0.001     34\n",
       "12  0.255919  0.135069  0.001     13\n",
       "40  0.256263  0.045249  0.001     41\n",
       "25  0.256528  0.070083  0.001     26\n",
       "39  0.258491  0.046385  0.001     40\n",
       "36  0.258976  0.047897  0.001     37\n",
       "34  0.259441  0.048968  0.001     35\n",
       "37  0.261749  0.047258  0.001     38\n",
       "28  0.262236  0.060982  0.001     29\n",
       "1   0.263404  0.316812  0.001      2\n",
       "38  0.264150  0.045217  0.001     39\n",
       "6   0.272420  0.194994  0.001      7\n",
       "41  0.275882  0.040384  0.001     42\n",
       "35  0.279350  0.050247  0.001     36\n",
       "7   0.285503  0.182704  0.001      8\n",
       "5   0.295849  0.209872  0.001      6\n",
       "3   0.297089  0.240364  0.001      4\n",
       "0   0.333004  0.520144  0.001      1"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Validation loss epochs in descending order\n",
    "history_df.drop(columns=['val_acc', 'acc']).sort_values(by='val_loss', ascending=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We also note that validation loss was also at an absolute minimum at epoch 14, so here is likely where the model begins to overfit."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "toc-hr-collapsed": true
   },
   "source": [
    "## Tuning"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Sagemaker automatic model tuning"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll use Sagemaker's built-in hyperparameter optimization to try to find a model \n",
    "with better validation accuracy. We'll use the (default) Bayesian strategy to search the hyperparameter space efficiently."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# architecture hyperparameter spaces\n",
    "conv0_hps = {'conv0_pad': IntegerParameter(1, 3),\n",
    "             'conv0_channels': IntegerParameter(24, 32),\n",
    "             'conv0_filter': IntegerParameter(2, 4),\n",
    "             'conv0_stride': IntegerParameter(1, 3),\n",
    "             'conv0_pool': IntegerParameter(1, 3),\n",
    "            }\n",
    "conv1_hps = {'conv1_pad': IntegerParameter(1, 3),\n",
    "             'conv1_channels': IntegerParameter(48, 64),\n",
    "             'conv1_filter': IntegerParameter(2, 4),\n",
    "             'conv1_stride': IntegerParameter(1, 3),\n",
    "             'conv1_pool': IntegerParameter(1, 3),\n",
    "            }\n",
    "conv2_hps = {'conv2_pad': IntegerParameter(1, 3),\n",
    "             'conv2_channels': IntegerParameter(96, 128),\n",
    "             'conv2_filter': IntegerParameter(2, 4),\n",
    "             'conv2_stride': IntegerParameter(1, 3),\n",
    "             'conv2_pool': IntegerParameter(1, 3),\n",
    "            }\n",
    "fc0_hps = {'fc0_neurons': IntegerParameter(200, 300)}\n",
    "fc1_hps = {'fc1_neurons': IntegerParameter(200, 300)}\n",
    "\n",
    "hyperparameter_ranges = {**conv0_hps, **conv1_hps, **conv2_hps, **fc0_hps, **fc1_hps}\n",
    "\n",
    "# objective and metric\n",
    "objective_metric_name = 'val_acc'\n",
    "objective_type = 'Maximize'\n",
    "metric_definitions = [{'Name': 'val_acc',\n",
    "                       'Regex': 'best_val_acc: ([0-9\\\\.]+)'}]\n",
    "\n",
    "# tuner\n",
    "tuner = HyperparameterTuner(tf_estimator,\n",
    "                            objective_metric_name,\n",
    "                            hyperparameter_ranges,\n",
    "                            metric_definitions,\n",
    "                            max_jobs=10,\n",
    "                            max_parallel_jobs=1,\n",
    "                            objective_type=objective_type)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "tuner.fit(paths)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Analyze tuning job results "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>conv0_channels</th>\n",
       "      <th>conv0_filter</th>\n",
       "      <th>conv0_pad</th>\n",
       "      <th>conv0_pool</th>\n",
       "      <th>conv0_stride</th>\n",
       "      <th>conv1_channels</th>\n",
       "      <th>conv1_filter</th>\n",
       "      <th>conv1_pad</th>\n",
       "      <th>conv1_pool</th>\n",
       "      <th>conv1_stride</th>\n",
       "      <th>...</th>\n",
       "      <th>conv2_pool</th>\n",
       "      <th>conv2_stride</th>\n",
       "      <th>fc0_neurons</th>\n",
       "      <th>fc1_neurons</th>\n",
       "      <th>TrainingJobName</th>\n",
       "      <th>TrainingJobStatus</th>\n",
       "      <th>FinalObjectiveValue</th>\n",
       "      <th>TrainingStartTime</th>\n",
       "      <th>TrainingEndTime</th>\n",
       "      <th>TrainingElapsedTimeSeconds</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>24.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>64.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>216.0</td>\n",
       "      <td>267.0</td>\n",
       "      <td>tensorflow-training-190920-1614-010-2339cf4b</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.9104</td>\n",
       "      <td>2019-09-20 17:23:39-07:00</td>\n",
       "      <td>2019-09-20 17:33:02-07:00</td>\n",
       "      <td>563.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>29.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>54.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>261.0</td>\n",
       "      <td>218.0</td>\n",
       "      <td>tensorflow-training-190920-1614-009-61a0d8ce</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.9235</td>\n",
       "      <td>2019-09-20 17:11:08-07:00</td>\n",
       "      <td>2019-09-20 17:20:00-07:00</td>\n",
       "      <td>532.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>26.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>55.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>204.0</td>\n",
       "      <td>233.0</td>\n",
       "      <td>tensorflow-training-190920-1614-008-98054b92</td>\n",
       "      <td>Failed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2019-09-20 17:07:44-07:00</td>\n",
       "      <td>2019-09-20 17:08:55-07:00</td>\n",
       "      <td>71.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>26.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>55.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>205.0</td>\n",
       "      <td>233.0</td>\n",
       "      <td>tensorflow-training-190920-1614-007-075957e0</td>\n",
       "      <td>Failed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2019-09-20 17:04:15-07:00</td>\n",
       "      <td>2019-09-20 17:05:29-07:00</td>\n",
       "      <td>74.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>27.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>63.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>225.0</td>\n",
       "      <td>300.0</td>\n",
       "      <td>tensorflow-training-190920-1614-006-b2bfc6ce</td>\n",
       "      <td>Failed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2019-09-20 17:00:31-07:00</td>\n",
       "      <td>2019-09-20 17:01:49-07:00</td>\n",
       "      <td>78.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>27.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>63.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>224.0</td>\n",
       "      <td>299.0</td>\n",
       "      <td>tensorflow-training-190920-1614-005-f7d4ee53</td>\n",
       "      <td>Failed</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2019-09-20 16:56:32-07:00</td>\n",
       "      <td>2019-09-20 16:58:06-07:00</td>\n",
       "      <td>94.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>32.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>58.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>253.0</td>\n",
       "      <td>234.0</td>\n",
       "      <td>tensorflow-training-190920-1614-004-527c5a6e</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.9057</td>\n",
       "      <td>2019-09-20 16:50:07-07:00</td>\n",
       "      <td>2019-09-20 16:54:22-07:00</td>\n",
       "      <td>255.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>28.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>48.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>...</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>249.0</td>\n",
       "      <td>242.0</td>\n",
       "      <td>tensorflow-training-190920-1614-003-9198b56d</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.8105</td>\n",
       "      <td>2019-09-20 16:36:50-07:00</td>\n",
       "      <td>2019-09-20 16:46:35-07:00</td>\n",
       "      <td>585.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>27.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>51.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>275.0</td>\n",
       "      <td>271.0</td>\n",
       "      <td>tensorflow-training-190920-1614-002-eb3e96e1</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.8946</td>\n",
       "      <td>2019-09-20 16:27:11-07:00</td>\n",
       "      <td>2019-09-20 16:33:21-07:00</td>\n",
       "      <td>370.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>31.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>57.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>...</td>\n",
       "      <td>2.0</td>\n",
       "      <td>3.0</td>\n",
       "      <td>214.0</td>\n",
       "      <td>279.0</td>\n",
       "      <td>tensorflow-training-190920-1614-001-f2b7ac23</td>\n",
       "      <td>Completed</td>\n",
       "      <td>0.8983</td>\n",
       "      <td>2019-09-20 16:16:17-07:00</td>\n",
       "      <td>2019-09-20 16:23:07-07:00</td>\n",
       "      <td>410.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>10 rows × 23 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   conv0_channels  conv0_filter  conv0_pad  conv0_pool  conv0_stride  \\\n",
       "0            24.0           3.0        1.0         2.0           3.0   \n",
       "1            29.0           3.0        3.0         1.0           2.0   \n",
       "2            26.0           2.0        2.0         1.0           3.0   \n",
       "3            26.0           2.0        2.0         1.0           3.0   \n",
       "4            27.0           4.0        2.0         3.0           1.0   \n",
       "5            27.0           4.0        2.0         3.0           1.0   \n",
       "6            32.0           2.0        3.0         1.0           2.0   \n",
       "7            28.0           2.0        3.0         3.0           2.0   \n",
       "8            27.0           2.0        2.0         2.0           2.0   \n",
       "9            31.0           2.0        1.0         2.0           2.0   \n",
       "\n",
       "   conv1_channels  conv1_filter  conv1_pad  conv1_pool  conv1_stride  ...  \\\n",
       "0            64.0           2.0        1.0         1.0           1.0  ...   \n",
       "1            54.0           4.0        3.0         3.0           1.0  ...   \n",
       "2            55.0           2.0        2.0         3.0           2.0  ...   \n",
       "3            55.0           2.0        2.0         3.0           2.0  ...   \n",
       "4            63.0           4.0        1.0         1.0           3.0  ...   \n",
       "5            63.0           4.0        1.0         1.0           3.0  ...   \n",
       "6            58.0           4.0        3.0         1.0           3.0  ...   \n",
       "7            48.0           2.0        1.0         1.0           3.0  ...   \n",
       "8            51.0           3.0        3.0         2.0           2.0  ...   \n",
       "9            57.0           4.0        2.0         1.0           2.0  ...   \n",
       "\n",
       "   conv2_pool  conv2_stride  fc0_neurons  fc1_neurons  \\\n",
       "0         1.0           2.0        216.0        267.0   \n",
       "1         1.0           1.0        261.0        218.0   \n",
       "2         3.0           3.0        204.0        233.0   \n",
       "3         3.0           3.0        205.0        233.0   \n",
       "4         3.0           3.0        225.0        300.0   \n",
       "5         3.0           3.0        224.0        299.0   \n",
       "6         3.0           3.0        253.0        234.0   \n",
       "7         3.0           2.0        249.0        242.0   \n",
       "8         1.0           2.0        275.0        271.0   \n",
       "9         2.0           3.0        214.0        279.0   \n",
       "\n",
       "                                TrainingJobName  TrainingJobStatus  \\\n",
       "0  tensorflow-training-190920-1614-010-2339cf4b          Completed   \n",
       "1  tensorflow-training-190920-1614-009-61a0d8ce          Completed   \n",
       "2  tensorflow-training-190920-1614-008-98054b92             Failed   \n",
       "3  tensorflow-training-190920-1614-007-075957e0             Failed   \n",
       "4  tensorflow-training-190920-1614-006-b2bfc6ce             Failed   \n",
       "5  tensorflow-training-190920-1614-005-f7d4ee53             Failed   \n",
       "6  tensorflow-training-190920-1614-004-527c5a6e          Completed   \n",
       "7  tensorflow-training-190920-1614-003-9198b56d          Completed   \n",
       "8  tensorflow-training-190920-1614-002-eb3e96e1          Completed   \n",
       "9  tensorflow-training-190920-1614-001-f2b7ac23          Completed   \n",
       "\n",
       "   FinalObjectiveValue         TrainingStartTime           TrainingEndTime  \\\n",
       "0               0.9104 2019-09-20 17:23:39-07:00 2019-09-20 17:33:02-07:00   \n",
       "1               0.9235 2019-09-20 17:11:08-07:00 2019-09-20 17:20:00-07:00   \n",
       "2                  NaN 2019-09-20 17:07:44-07:00 2019-09-20 17:08:55-07:00   \n",
       "3                  NaN 2019-09-20 17:04:15-07:00 2019-09-20 17:05:29-07:00   \n",
       "4                  NaN 2019-09-20 17:00:31-07:00 2019-09-20 17:01:49-07:00   \n",
       "5                  NaN 2019-09-20 16:56:32-07:00 2019-09-20 16:58:06-07:00   \n",
       "6               0.9057 2019-09-20 16:50:07-07:00 2019-09-20 16:54:22-07:00   \n",
       "7               0.8105 2019-09-20 16:36:50-07:00 2019-09-20 16:46:35-07:00   \n",
       "8               0.8946 2019-09-20 16:27:11-07:00 2019-09-20 16:33:21-07:00   \n",
       "9               0.8983 2019-09-20 16:16:17-07:00 2019-09-20 16:23:07-07:00   \n",
       "\n",
       "   TrainingElapsedTimeSeconds  \n",
       "0                       563.0  \n",
       "1                       532.0  \n",
       "2                        71.0  \n",
       "3                        74.0  \n",
       "4                        78.0  \n",
       "5                        94.0  \n",
       "6                       255.0  \n",
       "7                       585.0  \n",
       "8                       370.0  \n",
       "9                       410.0  \n",
       "\n",
       "[10 rows x 23 columns]"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# tuning job results dataframe\n",
    "tuning_job_df = tuner.analytics().dataframe()\n",
    "tuning_job_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    Completed\n",
       "1    Completed\n",
       "2       Failed\n",
       "3       Failed\n",
       "4       Failed\n",
       "5       Failed\n",
       "6    Completed\n",
       "7    Completed\n",
       "8    Completed\n",
       "9    Completed\n",
       "Name: TrainingJobStatus, dtype: object"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tuning_job_df['TrainingJobStatus']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We note that 4 out of 10 of the jobs failed. After inspecting the [CloudWatch job logs](console.aws.amazon.com/cloudwatch/home), we found that this was due to inappropropriate hyperparameter range choices leading to negative dimension errors. \n",
    "\n",
    "This seems especially problematic if Bayesian optimization is driving the search towards incompatible values of the hyperparameters -- training jobs would be more likely to fail and it would be harder to leave a region of hyperparameter space where such failure is likely. Greater care should be taken to avoide incompatible choices of hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1    0.9235\n",
       "0    0.9104\n",
       "6    0.9057\n",
       "9    0.8983\n",
       "8    0.8946\n",
       "7    0.8105\n",
       "2       NaN\n",
       "3       NaN\n",
       "4       NaN\n",
       "5       NaN\n",
       "Name: FinalObjectiveValue, dtype: float64"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tuning_job_df['FinalObjectiveValue'].sort_values(ascending=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Although the validation accuracy improved from job to job, none of the models thus trained achieved a validation accuracy better than the default model,"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the end, the default model hyperparameters seemed to be a good option. We'll check the test set performance of the sequence of models learned during [that training job](#Run-a-single-training-job).\n",
    "\n",
    "As [previously observed](#Analyze-training-history), we expect that weights from this period will perform best on test data, and will be a sound choice for a final model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "FashionMNISTCNN-epoch-01-val_acc-0.8778.hdf5\n",
      "FashionMNISTCNN-epoch-02-val_acc-0.9033.hdf5\n",
      "FashionMNISTCNN-epoch-03-val_acc-0.9091.hdf5\n",
      "FashionMNISTCNN-epoch-09-val_acc-0.9281.hdf5\n",
      "FashionMNISTCNN-epoch-12-val_acc-0.9281.hdf5\n",
      "FashionMNISTCNN-epoch-15-val_acc-0.9306.hdf5\n",
      "FashionMNISTCNN-epoch-16-val_acc-0.9310.hdf5\n",
      "FashionMNISTCNN-epoch-19-val_acc-0.9339.hdf5\n",
      "FashionMNISTCNN-epoch-24-val_acc-0.9367.hdf5\n",
      "FashionMNISTCNN-epoch-32-val_acc-0.9386.hdf5\n",
      "FashionMNISTCNN-history.csv\n",
      "dummy.txt\n"
     ]
    }
   ],
   "source": [
    "! ls models/keras_checkpoints"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll evaluate all models between epochs 15-32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def epoch_and_val_acc_from_file_name(model_file):\n",
    "    model_file = model_file.lstrip('FashionMNISTCNN-')\n",
    "    model_file = model_file.rstrip('.hdf5')\n",
    "    model_file = model_file.split('-')\n",
    "    epoch = int(model_file[1])\n",
    "    val_acc = float(model_file[3])\n",
    "    return epoch, val_acc\n",
    "\n",
    "def get_models_from_dir(model_dir, epoch_range, input_shape=(28, 28, 1), drop_rate=0.50):\n",
    "    models = {}\n",
    "    for _, _, model_files in os.walk(model_dir):\n",
    "        for model_file in sorted(model_files):\n",
    "            if '.hdf5' in model_file:\n",
    "                epoch, val_acc = epoch_and_val_acc_from_file_name(model_file)\n",
    "                if epoch in epoch_range:\n",
    "                    model = fmc(input_shape=input_shape, drop_rate=drop_rate)\n",
    "                    model.load_weights(os.path.join(model_dir, model_file))\n",
    "                    model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "                    models[epoch] = model\n",
    "    return models\n",
    "\n",
    "def model_eval_df(models, X, Y):\n",
    "    losses, accs = [], []\n",
    "    for epoch in models:\n",
    "        print(\"Evaluating epoch {} model:\\n\".format(epoch))\n",
    "        loss, acc = models[epoch].evaluate(x=X, y=Y)\n",
    "        losses += [loss]\n",
    "        accs += [acc]\n",
    "    eval_df = pd.DataFrame({'epoch': list(models.keys()), 'test_loss': losses, 'test_acc': accs})\n",
    "    return eval_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating epoch 15 model:\n",
      "\n",
      "10000/10000 [==============================] - 26s 3ms/step\n",
      "Evaluating epoch 16 model:\n",
      "\n",
      "10000/10000 [==============================] - 27s 3ms/step\n",
      "Evaluating epoch 19 model:\n",
      "\n",
      "10000/10000 [==============================] - 25s 3ms/step\n",
      "Evaluating epoch 24 model:\n",
      "\n",
      "10000/10000 [==============================] - 27s 3ms/step\n",
      "Evaluating epoch 32 model:\n",
      "\n",
      "10000/10000 [==============================] - 26s 3ms/step\n"
     ]
    }
   ],
   "source": [
    "# load and prepare test data\n",
    "(X_train, Y_train, X_val, Y_val, X_test, Y_test) = fmc.load_data()\n",
    "(X_train, Y_train, X_val, Y_val, X_test, Y_test) = fmc.prepare_data(X_train, Y_train, X_val, Y_val, X_test, Y_test)\n",
    "#evaluate models\n",
    "epoch_range = range(15, 33)\n",
    "models = get_models_from_dir('models/keras_checkpoints', epoch_range)\n",
    "model_test_eval_df = model_eval_df(models, X_test, Y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def plot_performance(model_df):\n",
    "    fig, ax = plt.subplots(1, 2, figsize=(15, 10))\n",
    "\n",
    "    plt.subplot(1, 2, 1)\n",
    "    plt.plot('epoch', 'loss', data=model_df, label='train_loss')\n",
    "    plt.plot('epoch', 'val_loss', data=model_df, label='val_loss')\n",
    "    plt.plot('epoch', 'test_loss', data=model_df, label='test_loss')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('Loss')\n",
    "    plt.title('Train, val and test loss')\n",
    "    plt.legend()\n",
    "    \n",
    "    plt.subplot(1, 2, 2)\n",
    "    plt.plot('epoch', 'acc', data=model_df, label='train_acc')\n",
    "    plt.plot('epoch', 'val_acc', data=model_df, label='val_acc')\n",
    "    plt.plot('epoch', 'test_acc', data=model_df, label='test_acc')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('Accuracy')\n",
    "    plt.title('Train, val and test accuracy')\n",
    "    plt.legend()\n",
    "    \n",
    "model_df = pd.merge(history_df, model_test_eval_df, on='epoch')\n",
    "plot_performance(model_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val_loss</th>\n",
       "      <th>val_acc</th>\n",
       "      <th>loss</th>\n",
       "      <th>acc</th>\n",
       "      <th>lr</th>\n",
       "      <th>epoch</th>\n",
       "      <th>test_loss</th>\n",
       "      <th>test_acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.214470</td>\n",
       "      <td>0.9339</td>\n",
       "      <td>0.099285</td>\n",
       "      <td>0.96342</td>\n",
       "      <td>0.001</td>\n",
       "      <td>19</td>\n",
       "      <td>0.198591</td>\n",
       "      <td>0.9389</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.231908</td>\n",
       "      <td>0.9367</td>\n",
       "      <td>0.076089</td>\n",
       "      <td>0.97210</td>\n",
       "      <td>0.001</td>\n",
       "      <td>24</td>\n",
       "      <td>0.213586</td>\n",
       "      <td>0.9386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.237604</td>\n",
       "      <td>0.9386</td>\n",
       "      <td>0.055796</td>\n",
       "      <td>0.97962</td>\n",
       "      <td>0.001</td>\n",
       "      <td>32</td>\n",
       "      <td>0.220641</td>\n",
       "      <td>0.9381</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.207615</td>\n",
       "      <td>0.9310</td>\n",
       "      <td>0.112607</td>\n",
       "      <td>0.95914</td>\n",
       "      <td>0.001</td>\n",
       "      <td>16</td>\n",
       "      <td>0.194638</td>\n",
       "      <td>0.9362</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.200692</td>\n",
       "      <td>0.9306</td>\n",
       "      <td>0.122466</td>\n",
       "      <td>0.95532</td>\n",
       "      <td>0.001</td>\n",
       "      <td>15</td>\n",
       "      <td>0.190690</td>\n",
       "      <td>0.9322</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   val_loss  val_acc      loss      acc     lr  epoch  test_loss  test_acc\n",
       "2  0.214470   0.9339  0.099285  0.96342  0.001     19   0.198591    0.9389\n",
       "3  0.231908   0.9367  0.076089  0.97210  0.001     24   0.213586    0.9386\n",
       "4  0.237604   0.9386  0.055796  0.97962  0.001     32   0.220641    0.9381\n",
       "1  0.207615   0.9310  0.112607  0.95914  0.001     16   0.194638    0.9362\n",
       "0  0.200692   0.9306  0.122466  0.95532  0.001     15   0.190690    0.9322"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# epochs ranked by test accuracy\n",
    "model_df.sort_values(by='test_acc', ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val_loss</th>\n",
       "      <th>val_acc</th>\n",
       "      <th>loss</th>\n",
       "      <th>acc</th>\n",
       "      <th>lr</th>\n",
       "      <th>epoch</th>\n",
       "      <th>test_loss</th>\n",
       "      <th>test_acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.200692</td>\n",
       "      <td>0.9306</td>\n",
       "      <td>0.122466</td>\n",
       "      <td>0.95532</td>\n",
       "      <td>0.001</td>\n",
       "      <td>15</td>\n",
       "      <td>0.190690</td>\n",
       "      <td>0.9322</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.207615</td>\n",
       "      <td>0.9310</td>\n",
       "      <td>0.112607</td>\n",
       "      <td>0.95914</td>\n",
       "      <td>0.001</td>\n",
       "      <td>16</td>\n",
       "      <td>0.194638</td>\n",
       "      <td>0.9362</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.214470</td>\n",
       "      <td>0.9339</td>\n",
       "      <td>0.099285</td>\n",
       "      <td>0.96342</td>\n",
       "      <td>0.001</td>\n",
       "      <td>19</td>\n",
       "      <td>0.198591</td>\n",
       "      <td>0.9389</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.231908</td>\n",
       "      <td>0.9367</td>\n",
       "      <td>0.076089</td>\n",
       "      <td>0.97210</td>\n",
       "      <td>0.001</td>\n",
       "      <td>24</td>\n",
       "      <td>0.213586</td>\n",
       "      <td>0.9386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.237604</td>\n",
       "      <td>0.9386</td>\n",
       "      <td>0.055796</td>\n",
       "      <td>0.97962</td>\n",
       "      <td>0.001</td>\n",
       "      <td>32</td>\n",
       "      <td>0.220641</td>\n",
       "      <td>0.9381</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   val_loss  val_acc      loss      acc     lr  epoch  test_loss  test_acc\n",
       "0  0.200692   0.9306  0.122466  0.95532  0.001     15   0.190690    0.9322\n",
       "1  0.207615   0.9310  0.112607  0.95914  0.001     16   0.194638    0.9362\n",
       "2  0.214470   0.9339  0.099285  0.96342  0.001     19   0.198591    0.9389\n",
       "3  0.231908   0.9367  0.076089  0.97210  0.001     24   0.213586    0.9386\n",
       "4  0.237604   0.9386  0.055796  0.97962  0.001     32   0.220641    0.9381"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# epochs ranked by test loss\n",
    "model_df.sort_values(by='test_loss', ascending=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As a compromise between test accuracy and loss, we'll select the epoch 19 model for the final model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Conclusions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We found that the default model architecture performed well with a test classifiction accuracy of $\\approx 93.9\\%$ and a categorical cross entropy loss of $\\approx 0.199$.\n",
    "\n",
    "Some possibilities for model improvement are:\n",
    "- Using data augmentation to increase the size of the training set. This is very easy to implement in Keras\n",
    "- Better hyperparameter tuning, particularly architecture parameters. This could be done by a more careful definition of the hyperparameter spaces used in [Bayesian tuning](#Analyze-tuning-job-results), or by random search nearby the default hyperparameters"
   ]
  }
 ],
 "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": "3.7.3"
  },
  "toc-autonumbering": false
 },
 "nbformat": 4,
 "nbformat_minor": 4
}