notebooks/tutorial/tutorial_quick.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Quick mcfly tutorial: RacketSports dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The goal of this tutorial is to get you familiar with training Neural Networks for time series using mcfly. At the end of the tutorial, you will have compared several Neural Network architectures you know how to train the best performing network.\n",
    "\n",
    "Here we use a rather simple time seris dataset as an example, the publicly available [RacketSports dataset](http://www.timeseriesclassification.com/description.php?Dataset=RacketSports). It contains time series data from movement sensors from mobile phones when playing either Squash or Badminton. The data is labelled with the activity types that these individuals did and the aim is to train and evaluate a *classifier*.\n",
    "\n",
    "Before you can start, please make sure you install mcfly (see the [mcfly installation page](https://github.com/NLeSC/mcfly))."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import required Python modules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "import sys\n",
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "# mcfly\n",
    "import mcfly\n",
    "import tensorflow as tf\n",
    "np.random.seed(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "sys.path.insert(0, os.path.abspath('../..'))\n",
    "from utils import tutorial_racketsports"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Download data pre-processed data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We have created a function for you to fetch the preprocessed data from https://zenodo.org/record/3743603. Please specify the `directory_to_extract_to` in the code below and then execute the cell. This will download the preprocessed data into the directory in the `RacketSports` subdirectory. The output of the function is the path where the preprocessed data was stored."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Specify in which directory you want to store the data:\n",
    "directory_to_extract_to = '.'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data...\n",
      "Extracting data...\n",
      "Done\n"
     ]
    }
   ],
   "source": [
    "data_path = tutorial_racketsports.download_preprocessed_data(directory_to_extract_to)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A bit about the data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The [RacketSports dataset](http://www.timeseriesclassification.com/description.php?Dataset=RacketSports) contains motion sensor data of a smart watch from people playing Squash or Badminton. It contains the sensor data from gyroscope (= 3 channels) and accelerometer (= 3 channels), hence has a total of 6 channels. \n",
    "\n",
    "The problem is to identify which sport and which stroke the players are making. The data was collected at a rate of 10 HZ over 3 seconds whilst the player played either a forehand/backhand in squash or a clear/smash in badminton.\n",
    "\n",
    "The goal of classification is to assign an activity label to a previously unseen segment."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the pre-processed data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load the preprocessed data as stored in Numpy-files. Please note that the data has already been split up into a training (\\~50%), validation (\\~25%), and test (\\~25%) subsets. It is common practice to call the input data X and the labels y."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, y_train, X_val, y_val, X_test, y_test = tutorial_racketsports.load_data(data_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Data X and labels y are of type Numpy array. In the cell below we inspect the shape of the data. As you can see the shape of X is expressed as a Python tuple containing: the number of samples, length of the time series, and the number of channels for each sample. Similarly, the shape of y is represents the number of samples and the number of classes (unique labels). "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X training shape: (151, 30, 6)\n",
      "y training shape: (151,)\n",
      "y validation shape: (75,)\n",
      "y test shape: (77,)\n"
     ]
    }
   ],
   "source": [
    "print('X training shape:', X_train.shape)\n",
    "print('y training shape:', y_train.shape)\n",
    "print('y validation shape:', y_val.shape)\n",
    "print('y test shape:', y_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inspect labels (and prepare for model training)\n",
    "The labels are given as strings as provided in the original data. Those indicate one of the four possible classes:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Badminton_Clear' 'Badminton_Smash' 'Squash_BackhandBoast'\n",
      " 'Squash_ForehandBoast']\n"
     ]
    }
   ],
   "source": [
    "labels_unique = np.unique(y_train)\n",
    "print(labels_unique)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create binary labels\n",
    "To train a classifier the labels need to be translated into binary labels (so-called **one-hot encoding**)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>#sk-container-id-1 {color: black;background-color: white;}#sk-container-id-1 pre{padding: 0;}#sk-container-id-1 div.sk-toggleable {background-color: white;}#sk-container-id-1 label.sk-toggleable__label {cursor: pointer;display: block;width: 100%;margin-bottom: 0;padding: 0.3em;box-sizing: border-box;text-align: center;}#sk-container-id-1 label.sk-toggleable__label-arrow:before {content: \"▸\";float: left;margin-right: 0.25em;color: #696969;}#sk-container-id-1 label.sk-toggleable__label-arrow:hover:before {color: black;}#sk-container-id-1 div.sk-estimator:hover label.sk-toggleable__label-arrow:before {color: black;}#sk-container-id-1 div.sk-toggleable__content {max-height: 0;max-width: 0;overflow: hidden;text-align: left;background-color: #f0f8ff;}#sk-container-id-1 div.sk-toggleable__content pre {margin: 0.2em;color: black;border-radius: 0.25em;background-color: #f0f8ff;}#sk-container-id-1 input.sk-toggleable__control:checked~div.sk-toggleable__content {max-height: 200px;max-width: 100%;overflow: auto;}#sk-container-id-1 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {content: \"▾\";}#sk-container-id-1 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 input.sk-hidden--visually {border: 0;clip: rect(1px 1px 1px 1px);clip: rect(1px, 1px, 1px, 1px);height: 1px;margin: -1px;overflow: hidden;padding: 0;position: absolute;width: 1px;}#sk-container-id-1 div.sk-estimator {font-family: monospace;background-color: #f0f8ff;border: 1px dotted black;border-radius: 0.25em;box-sizing: border-box;margin-bottom: 0.5em;}#sk-container-id-1 div.sk-estimator:hover {background-color: #d4ebff;}#sk-container-id-1 div.sk-parallel-item::after {content: \"\";width: 100%;border-bottom: 1px solid gray;flex-grow: 1;}#sk-container-id-1 div.sk-label:hover label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-serial::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: 0;}#sk-container-id-1 div.sk-serial {display: flex;flex-direction: column;align-items: center;background-color: white;padding-right: 0.2em;padding-left: 0.2em;position: relative;}#sk-container-id-1 div.sk-item {position: relative;z-index: 1;}#sk-container-id-1 div.sk-parallel {display: flex;align-items: stretch;justify-content: center;background-color: white;position: relative;}#sk-container-id-1 div.sk-item::before, #sk-container-id-1 div.sk-parallel-item::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: -1;}#sk-container-id-1 div.sk-parallel-item {display: flex;flex-direction: column;z-index: 1;position: relative;background-color: white;}#sk-container-id-1 div.sk-parallel-item:first-child::after {align-self: flex-end;width: 50%;}#sk-container-id-1 div.sk-parallel-item:last-child::after {align-self: flex-start;width: 50%;}#sk-container-id-1 div.sk-parallel-item:only-child::after {width: 0;}#sk-container-id-1 div.sk-dashed-wrapped {border: 1px dashed gray;margin: 0 0.4em 0.5em 0.4em;box-sizing: border-box;padding-bottom: 0.4em;background-color: white;}#sk-container-id-1 div.sk-label label {font-family: monospace;font-weight: bold;display: inline-block;line-height: 1.2em;}#sk-container-id-1 div.sk-label-container {text-align: center;}#sk-container-id-1 div.sk-container {/* jupyter's `normalize.less` sets `[hidden] { display: none; }` but bootstrap.min.css set `[hidden] { display: none !important; }` so we also need the `!important` here to be able to override the default hidden behavior on the sphinx rendered scikit-learn.org. See: https://github.com/scikit-learn/scikit-learn/issues/21755 */display: inline-block !important;position: relative;}#sk-container-id-1 div.sk-text-repr-fallback {display: none;}</style><div id=\"sk-container-id-1\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>LabelBinarizer()</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-1\" type=\"checkbox\" checked><label for=\"sk-estimator-id-1\" class=\"sk-toggleable__label sk-toggleable__label-arrow\">LabelBinarizer</label><div class=\"sk-toggleable__content\"><pre>LabelBinarizer()</pre></div></div></div></div></div>"
      ],
      "text/plain": [
       "LabelBinarizer()"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn import preprocessing\n",
    "lb = preprocessing.LabelBinarizer()\n",
    "lb.fit(labels_unique)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train_binary = lb.fit_transform(y_train)\n",
    "y_val_binary = lb.fit_transform(y_val)\n",
    "y_test_binary = lb.fit_transform(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1, 0, 0, 0],\n",
       "       [0, 0, 0, 1],\n",
       "       [0, 0, 1, 0],\n",
       "       [0, 1, 0, 0],\n",
       "       [1, 0, 0, 0],\n",
       "       [0, 0, 1, 0],\n",
       "       [0, 0, 1, 0],\n",
       "       [0, 0, 1, 0],\n",
       "       [0, 0, 1, 0],\n",
       "       [1, 0, 0, 0]])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train_binary[:10,:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If we want to see the class labels again, we can use:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Badminton_Clear' 'Badminton_Smash' 'Squash_BackhandBoast'\n",
      " 'Squash_ForehandBoast']\n"
     ]
    }
   ],
   "source": [
    "print(lb.classes_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize the data \n",
    "(optional, but usually nice to get an idea of how it looks like and if everything is as expected)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x272623d2590>"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "plt.style.use('ggplot')\n",
    "plt.plot(np.array(X_train[10,:,:]))\n",
    "plt.xlabel('time points')\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *Question 1: How many channels does this dataset have?*\n",
    "### *Question 2: What is the least common activity label in this dataset?*\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First step in the development of any deep learning model is to create a model architecture. As we do not know what architecture is best for our data we will create a set of random models to investigate which architecture is most suitable for our data and classification task. This process, creating random models, checking how good they are and then selecting the best one is called a 'random search'. A random search is considered to be the most robust approach to finding a good model. You will need to specificy how many models you want to create with argument 'number_of_models'. See for a full overview of the optional arguments the function documentation of modelgen.generate_models by running `modelgen.generate_models?`.\n",
    "\n",
    "##### What number of models to select?\n",
    "This number differs per dataset. More models will give better results but it will take longer to evaluate them. Because mcfly uses random search, you will get better results when using more models. Here we will generate 8 models to keep the training short for this tutorial, but it could just as well be done for 12, or 20 models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The value of metrics is set from ['accuracy'] (default) to ['accuracy']\n",
      "Set maximum kernel size for InceptionTime models to number of timesteps.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n",
      "WARNING:absl:`lr` is deprecated, please use `learning_rate` instead, or use the legacy optimizer, e.g.,tf.keras.optimizers.legacy.Adam.\n"
     ]
    }
   ],
   "source": [
    "num_classes = y_train_binary.shape[1]\n",
    "metric = 'accuracy'\n",
    "models = mcfly.modelgen.generate_models(X_train.shape,\n",
    "                                        number_of_output_dimensions=num_classes,\n",
    "                                        number_of_models = 8,\n",
    "                                        metrics=[metric])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inspect the models\n",
    "We can have a look at the models that were generated. The layers are shown as table rows. Most common layer types are 'Convolution' and 'LSTM' and 'Dense'. For more information see the [mcfly user manual](https://github.com/NLeSC/mcfly/wiki/User-manual) and the [tutorial cheat sheet](https://github.com/NLeSC/mcfly-tutorial/blob/master/cheatsheet.md). The summary also shows the data shape of each layer output and the number of parameters that are trained within this layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 0 \n",
      "\n",
      "Model type: ResNet \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.005481464347736465, 'regularization_rate': 0.010209289921137055, 'network_depth': 5, 'min_filters_number': 107, 'max_kernel_size': 29}\n",
      " \n",
      "Model description:\n",
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_1 (InputLayer)           [(None, 30, 6)]      0           []                               \n",
      "                                                                                                  \n",
      " batch_normalization (BatchNorm  (None, 30, 6)       24          ['input_1[0][0]']                \n",
      " alization)                                                                                       \n",
      "                                                                                                  \n",
      " conv1d (Conv1D)                (None, 30, 107)      18725       ['batch_normalization[0][0]']    \n",
      "                                                                                                  \n",
      " batch_normalization_1 (BatchNo  (None, 30, 107)     428         ['conv1d[0][0]']                 \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu (ReLU)                   (None, 30, 107)      0           ['batch_normalization_1[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_1 (Conv1D)              (None, 30, 107)      332128      ['re_lu[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_2 (BatchNo  (None, 30, 107)     428         ['conv1d_1[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_1 (ReLU)                 (None, 30, 107)      0           ['batch_normalization_2[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_2 (Conv1D)              (None, 30, 107)      332128      ['re_lu_1[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_3 (BatchNo  (None, 30, 107)     428         ['conv1d_2[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_2 (ReLU)                 (None, 30, 107)      0           ['batch_normalization_3[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_3 (Conv1D)              (None, 30, 107)      11556       ['re_lu_2[0][0]']                \n",
      "                                                                                                  \n",
      " add (Add)                      (None, 30, 107)      0           ['re_lu_2[0][0]',                \n",
      "                                                                  'conv1d_3[0][0]']               \n",
      "                                                                                                  \n",
      " conv1d_4 (Conv1D)              (None, 30, 150)      321150      ['add[0][0]']                    \n",
      "                                                                                                  \n",
      " batch_normalization_4 (BatchNo  (None, 30, 150)     600         ['conv1d_4[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_3 (ReLU)                 (None, 30, 150)      0           ['batch_normalization_4[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_5 (Conv1D)              (None, 30, 150)      450150      ['re_lu_3[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_5 (BatchNo  (None, 30, 150)     600         ['conv1d_5[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_4 (ReLU)                 (None, 30, 150)      0           ['batch_normalization_5[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_6 (Conv1D)              (None, 30, 150)      450150      ['re_lu_4[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_6 (BatchNo  (None, 30, 150)     600         ['conv1d_6[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_5 (ReLU)                 (None, 30, 150)      0           ['batch_normalization_6[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_7 (Conv1D)              (None, 30, 150)      22650       ['re_lu_5[0][0]']                \n",
      "                                                                                                  \n",
      " add_1 (Add)                    (None, 30, 150)      0           ['re_lu_5[0][0]',                \n",
      "                                                                  'conv1d_7[0][0]']               \n",
      "                                                                                                  \n",
      " conv1d_8 (Conv1D)              (None, 30, 212)      445412      ['add_1[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_7 (BatchNo  (None, 30, 212)     848         ['conv1d_8[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_6 (ReLU)                 (None, 30, 212)      0           ['batch_normalization_7[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_9 (Conv1D)              (None, 30, 212)      629428      ['re_lu_6[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_8 (BatchNo  (None, 30, 212)     848         ['conv1d_9[0][0]']               \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_7 (ReLU)                 (None, 30, 212)      0           ['batch_normalization_8[0][0]']  \n",
      "                                                                                                  \n",
      " conv1d_10 (Conv1D)             (None, 30, 212)      629428      ['re_lu_7[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_9 (BatchNo  (None, 30, 212)     848         ['conv1d_10[0][0]']              \n",
      " rmalization)                                                                                     \n",
      "                                                                                                  \n",
      " re_lu_8 (ReLU)                 (None, 30, 212)      0           ['batch_normalization_9[0][0]']  \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                                                  \n",
      " conv1d_11 (Conv1D)             (None, 30, 212)      45156       ['re_lu_8[0][0]']                \n",
      "                                                                                                  \n",
      " add_2 (Add)                    (None, 30, 212)      0           ['re_lu_8[0][0]',                \n",
      "                                                                  'conv1d_11[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_12 (Conv1D)             (None, 30, 299)      634179      ['add_2[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_10 (BatchN  (None, 30, 299)     1196        ['conv1d_12[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_9 (ReLU)                 (None, 30, 299)      0           ['batch_normalization_10[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_13 (Conv1D)             (None, 30, 299)      894309      ['re_lu_9[0][0]']                \n",
      "                                                                                                  \n",
      " batch_normalization_11 (BatchN  (None, 30, 299)     1196        ['conv1d_13[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_10 (ReLU)                (None, 30, 299)      0           ['batch_normalization_11[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_14 (Conv1D)             (None, 30, 299)      894309      ['re_lu_10[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_12 (BatchN  (None, 30, 299)     1196        ['conv1d_14[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_11 (ReLU)                (None, 30, 299)      0           ['batch_normalization_12[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_15 (Conv1D)             (None, 30, 299)      89700       ['re_lu_11[0][0]']               \n",
      "                                                                                                  \n",
      " add_3 (Add)                    (None, 30, 299)      0           ['re_lu_11[0][0]',               \n",
      "                                                                  'conv1d_15[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_16 (Conv1D)             (None, 30, 422)      883668      ['add_3[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_13 (BatchN  (None, 30, 422)     1688        ['conv1d_16[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_12 (ReLU)                (None, 30, 422)      0           ['batch_normalization_13[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_17 (Conv1D)             (None, 30, 422)      1247010     ['re_lu_12[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_14 (BatchN  (None, 30, 422)     1688        ['conv1d_17[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_13 (ReLU)                (None, 30, 422)      0           ['batch_normalization_14[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_18 (Conv1D)             (None, 30, 422)      1247010     ['re_lu_13[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_15 (BatchN  (None, 30, 422)     1688        ['conv1d_18[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_14 (ReLU)                (None, 30, 422)      0           ['batch_normalization_15[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_19 (Conv1D)             (None, 30, 422)      178506      ['re_lu_14[0][0]']               \n",
      "                                                                                                  \n",
      " add_4 (Add)                    (None, 30, 422)      0           ['re_lu_14[0][0]',               \n",
      "                                                                  'conv1d_19[0][0]']              \n",
      "                                                                                                  \n",
      " global_average_pooling1d (Glob  (None, 422)         0           ['add_4[0][0]']                  \n",
      " alAveragePooling1D)                                                                              \n",
      "                                                                                                  \n",
      " dense (Dense)                  (None, 4)            1692        ['global_average_pooling1d[0][0]'\n",
      "                                                                 ]                                \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 9,772,748\n",
      "Trainable params: 9,765,596\n",
      "Non-trainable params: 7,152\n",
      "__________________________________________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 1 \n",
      "\n",
      "Model type: InceptionTime \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.0436664988829779, 'regularization_rate': 0.003503259200059715, 'network_depth': 3, 'filters_number': 69, 'max_kernel_size': 17}\n",
      " \n",
      "Model description:\n",
      "Model: \"model_1\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_2 (InputLayer)           [(None, 30, 6)]      0           []                               \n",
      "                                                                                                  \n",
      " batch_normalization_16 (BatchN  (None, 30, 6)       24          ['input_2[0][0]']                \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " conv1d_20 (Conv1D)             (None, 30, 32)       192         ['batch_normalization_16[0][0]'] \n",
      "                                                                                                  \n",
      " max_pooling1d (MaxPooling1D)   (None, 30, 6)        0           ['batch_normalization_16[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_21 (Conv1D)             (None, 30, 69)       37536       ['conv1d_20[0][0]']              \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                                                  \n",
      " conv1d_22 (Conv1D)             (None, 30, 69)       17664       ['conv1d_20[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_23 (Conv1D)             (None, 30, 69)       8832        ['conv1d_20[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_24 (Conv1D)             (None, 30, 69)       414         ['max_pooling1d[0][0]']          \n",
      "                                                                                                  \n",
      " concatenate (Concatenate)      (None, 30, 276)      0           ['conv1d_21[0][0]',              \n",
      "                                                                  'conv1d_22[0][0]',              \n",
      "                                                                  'conv1d_23[0][0]',              \n",
      "                                                                  'conv1d_24[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_17 (BatchN  (None, 30, 276)     1104        ['concatenate[0][0]']            \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation (Activation)        (None, 30, 276)      0           ['batch_normalization_17[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_25 (Conv1D)             (None, 30, 32)       8832        ['activation[0][0]']             \n",
      "                                                                                                  \n",
      " max_pooling1d_1 (MaxPooling1D)  (None, 30, 276)     0           ['activation[0][0]']             \n",
      "                                                                                                  \n",
      " conv1d_26 (Conv1D)             (None, 30, 69)       37536       ['conv1d_25[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_27 (Conv1D)             (None, 30, 69)       17664       ['conv1d_25[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_28 (Conv1D)             (None, 30, 69)       8832        ['conv1d_25[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_29 (Conv1D)             (None, 30, 69)       19044       ['max_pooling1d_1[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_1 (Concatenate)    (None, 30, 276)      0           ['conv1d_26[0][0]',              \n",
      "                                                                  'conv1d_27[0][0]',              \n",
      "                                                                  'conv1d_28[0][0]',              \n",
      "                                                                  'conv1d_29[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_18 (BatchN  (None, 30, 276)     1104        ['concatenate_1[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_1 (Activation)      (None, 30, 276)      0           ['batch_normalization_18[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_30 (Conv1D)             (None, 30, 32)       8832        ['activation_1[0][0]']           \n",
      "                                                                                                  \n",
      " max_pooling1d_2 (MaxPooling1D)  (None, 30, 276)     0           ['activation_1[0][0]']           \n",
      "                                                                                                  \n",
      " conv1d_31 (Conv1D)             (None, 30, 69)       37536       ['conv1d_30[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_32 (Conv1D)             (None, 30, 69)       17664       ['conv1d_30[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_33 (Conv1D)             (None, 30, 69)       8832        ['conv1d_30[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_34 (Conv1D)             (None, 30, 69)       19044       ['max_pooling1d_2[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_2 (Concatenate)    (None, 30, 276)      0           ['conv1d_31[0][0]',              \n",
      "                                                                  'conv1d_32[0][0]',              \n",
      "                                                                  'conv1d_33[0][0]',              \n",
      "                                                                  'conv1d_34[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_35 (Conv1D)             (None, 30, 276)      1656        ['batch_normalization_16[0][0]'] \n",
      "                                                                                                  \n",
      " batch_normalization_19 (BatchN  (None, 30, 276)     1104        ['concatenate_2[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " batch_normalization_20 (BatchN  (None, 30, 276)     1104        ['conv1d_35[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_2 (Activation)      (None, 30, 276)      0           ['batch_normalization_19[0][0]'] \n",
      "                                                                                                  \n",
      " add_5 (Add)                    (None, 30, 276)      0           ['batch_normalization_20[0][0]', \n",
      "                                                                  'activation_2[0][0]']           \n",
      "                                                                                                  \n",
      " activation_3 (Activation)      (None, 30, 276)      0           ['add_5[0][0]']                  \n",
      "                                                                                                  \n",
      " global_average_pooling1d_1 (Gl  (None, 276)         0           ['activation_3[0][0]']           \n",
      " obalAveragePooling1D)                                                                            \n",
      "                                                                                                  \n",
      " dense_1 (Dense)                (None, 4)            1108        ['global_average_pooling1d_1[0][0\n",
      "                                                                 ]']                              \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 255,658\n",
      "Trainable params: 253,438\n",
      "Non-trainable params: 2,220\n",
      "__________________________________________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 2 \n",
      "\n",
      "Model type: DeepConvLSTM \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.0394693672382836, 'regularization_rate': 0.002879158913712458, 'filters': [48, 43, 68, 77], 'lstm_dims': [78]}\n",
      " \n",
      "Model description:\n",
      "Model: \"sequential\"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " batch_normalization_21 (Bat  (None, 30, 6)            24        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " reshape (Reshape)           (None, 30, 6, 1)          0         \n",
      "                                                                 \n",
      " conv2d (Conv2D)             (None, 30, 6, 48)         192       \n",
      "                                                                 \n",
      " batch_normalization_22 (Bat  (None, 30, 6, 48)        192       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_4 (Activation)   (None, 30, 6, 48)         0         \n",
      "                                                                 \n",
      " conv2d_1 (Conv2D)           (None, 30, 6, 43)         6235      \n",
      "                                                                 \n",
      " batch_normalization_23 (Bat  (None, 30, 6, 43)        172       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_5 (Activation)   (None, 30, 6, 43)         0         \n",
      "                                                                 \n",
      " conv2d_2 (Conv2D)           (None, 30, 6, 68)         8840      \n",
      "                                                                 \n",
      " batch_normalization_24 (Bat  (None, 30, 6, 68)        272       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_6 (Activation)   (None, 30, 6, 68)         0         \n",
      "                                                                 \n",
      " conv2d_3 (Conv2D)           (None, 30, 6, 77)         15785     \n",
      "                                                                 \n",
      " batch_normalization_25 (Bat  (None, 30, 6, 77)        308       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_7 (Activation)   (None, 30, 6, 77)         0         \n",
      "                                                                 \n",
      " reshape_1 (Reshape)         (None, 30, 462)           0         \n",
      "                                                                 \n",
      " lstm (LSTM)                 (None, 30, 78)            168792    \n",
      "                                                                 \n",
      " dropout (Dropout)           (None, 30, 78)            0         \n",
      "                                                                 \n",
      " time_distributed (TimeDistr  (None, 30, 4)            316       \n",
      " ibuted)                                                         \n",
      "                                                                 \n",
      " activation_8 (Activation)   (None, 30, 4)             0         \n",
      "                                                                 \n",
      " lambda (Lambda)             (None, 4)                 0         \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 201,128\n",
      "Trainable params: 200,644\n",
      "Non-trainable params: 484\n",
      "_________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 3 \n",
      "\n",
      "Model type: CNN \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.012851236876927944, 'regularization_rate': 0.013689688066052806, 'filters': array([ 41,  76,  90,  62,  86,  60,  14, 100,  73,  89]), 'fc_hidden_nodes': 1595}\n",
      " \n",
      "Model description:\n",
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " batch_normalization_26 (Bat  (None, 30, 6)            24        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " conv1d_36 (Conv1D)          (None, 30, 41)            779       \n",
      "                                                                 \n",
      " batch_normalization_27 (Bat  (None, 30, 41)           164       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_9 (Activation)   (None, 30, 41)            0         \n",
      "                                                                 \n",
      " conv1d_37 (Conv1D)          (None, 30, 76)            9424      \n",
      "                                                                 \n",
      " batch_normalization_28 (Bat  (None, 30, 76)           304       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_10 (Activation)  (None, 30, 76)            0         \n",
      "                                                                 \n",
      " conv1d_38 (Conv1D)          (None, 30, 90)            20610     \n",
      "                                                                 \n",
      " batch_normalization_29 (Bat  (None, 30, 90)           360       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_11 (Activation)  (None, 30, 90)            0         \n",
      "                                                                 \n",
      " conv1d_39 (Conv1D)          (None, 30, 62)            16802     \n",
      "                                                                 \n",
      " batch_normalization_30 (Bat  (None, 30, 62)           248       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_12 (Activation)  (None, 30, 62)            0         \n",
      "                                                                 \n",
      " conv1d_40 (Conv1D)          (None, 30, 86)            16082     \n",
      "                                                                 \n",
      " batch_normalization_31 (Bat  (None, 30, 86)           344       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_13 (Activation)  (None, 30, 86)            0         \n",
      "                                                                 \n",
      " conv1d_41 (Conv1D)          (None, 30, 60)            15540     \n",
      "                                                                 \n",
      " batch_normalization_32 (Bat  (None, 30, 60)           240       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_14 (Activation)  (None, 30, 60)            0         \n",
      "                                                                 \n",
      " conv1d_42 (Conv1D)          (None, 30, 14)            2534      \n",
      "                                                                 \n",
      " batch_normalization_33 (Bat  (None, 30, 14)           56        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_15 (Activation)  (None, 30, 14)            0         \n",
      "                                                                 \n",
      " conv1d_43 (Conv1D)          (None, 30, 100)           4300      \n",
      "                                                                 \n",
      " batch_normalization_34 (Bat  (None, 30, 100)          400       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_16 (Activation)  (None, 30, 100)           0         \n",
      "                                                                 \n",
      " conv1d_44 (Conv1D)          (None, 30, 73)            21973     \n",
      "                                                                 \n",
      " batch_normalization_35 (Bat  (None, 30, 73)           292       \n",
      " chNormalization)                                                \n",
      "                                                                 \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " activation_17 (Activation)  (None, 30, 73)            0         \n",
      "                                                                 \n",
      " conv1d_45 (Conv1D)          (None, 30, 89)            19580     \n",
      "                                                                 \n",
      " batch_normalization_36 (Bat  (None, 30, 89)           356       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_18 (Activation)  (None, 30, 89)            0         \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 2670)              0         \n",
      "                                                                 \n",
      " dense_3 (Dense)             (None, 1595)              4260245   \n",
      "                                                                 \n",
      " activation_19 (Activation)  (None, 1595)              0         \n",
      "                                                                 \n",
      " dense_4 (Dense)             (None, 4)                 6384      \n",
      "                                                                 \n",
      " batch_normalization_37 (Bat  (None, 4)                16        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_20 (Activation)  (None, 4)                 0         \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 4,397,057\n",
      "Trainable params: 4,395,655\n",
      "Non-trainable params: 1,402\n",
      "_________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 4 \n",
      "\n",
      "Model type: ResNet \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.008923210639822017, 'regularization_rate': 0.003950367924123254, 'network_depth': 4, 'min_filters_number': 47, 'max_kernel_size': 17}\n",
      " \n",
      "Model description:\n",
      "Model: \"model_2\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_3 (InputLayer)           [(None, 30, 6)]      0           []                               \n",
      "                                                                                                  \n",
      " batch_normalization_38 (BatchN  (None, 30, 6)       24          ['input_3[0][0]']                \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " conv1d_46 (Conv1D)             (None, 30, 47)       4841        ['batch_normalization_38[0][0]'] \n",
      "                                                                                                  \n",
      " batch_normalization_39 (BatchN  (None, 30, 47)      188         ['conv1d_46[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_15 (ReLU)                (None, 30, 47)       0           ['batch_normalization_39[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_47 (Conv1D)             (None, 30, 47)       37600       ['re_lu_15[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_40 (BatchN  (None, 30, 47)      188         ['conv1d_47[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_16 (ReLU)                (None, 30, 47)       0           ['batch_normalization_40[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_48 (Conv1D)             (None, 30, 47)       37600       ['re_lu_16[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_41 (BatchN  (None, 30, 47)      188         ['conv1d_48[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_17 (ReLU)                (None, 30, 47)       0           ['batch_normalization_41[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_49 (Conv1D)             (None, 30, 47)       2256        ['re_lu_17[0][0]']               \n",
      "                                                                                                  \n",
      " add_6 (Add)                    (None, 30, 47)       0           ['re_lu_17[0][0]',               \n",
      "                                                                  'conv1d_49[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_50 (Conv1D)             (None, 30, 66)       37290       ['add_6[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_42 (BatchN  (None, 30, 66)      264         ['conv1d_50[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_18 (ReLU)                (None, 30, 66)       0           ['batch_normalization_42[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_51 (Conv1D)             (None, 30, 66)       52338       ['re_lu_18[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_43 (BatchN  (None, 30, 66)      264         ['conv1d_51[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_19 (ReLU)                (None, 30, 66)       0           ['batch_normalization_43[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_52 (Conv1D)             (None, 30, 66)       52338       ['re_lu_19[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_44 (BatchN  (None, 30, 66)      264         ['conv1d_52[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_20 (ReLU)                (None, 30, 66)       0           ['batch_normalization_44[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_53 (Conv1D)             (None, 30, 66)       4422        ['re_lu_20[0][0]']               \n",
      "                                                                                                  \n",
      " add_7 (Add)                    (None, 30, 66)       0           ['re_lu_20[0][0]',               \n",
      "                                                                  'conv1d_53[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_54 (Conv1D)             (None, 30, 93)       49197       ['add_7[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_45 (BatchN  (None, 30, 93)      372         ['conv1d_54[0][0]']              \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_21 (ReLU)                (None, 30, 93)       0           ['batch_normalization_45[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_55 (Conv1D)             (None, 30, 93)       69285       ['re_lu_21[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_46 (BatchN  (None, 30, 93)      372         ['conv1d_55[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_22 (ReLU)                (None, 30, 93)       0           ['batch_normalization_46[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_56 (Conv1D)             (None, 30, 93)       69285       ['re_lu_22[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_47 (BatchN  (None, 30, 93)      372         ['conv1d_56[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_23 (ReLU)                (None, 30, 93)       0           ['batch_normalization_47[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_57 (Conv1D)             (None, 30, 93)       8742        ['re_lu_23[0][0]']               \n",
      "                                                                                                  \n",
      " add_8 (Add)                    (None, 30, 93)       0           ['re_lu_23[0][0]',               \n",
      "                                                                  'conv1d_57[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_58 (Conv1D)             (None, 30, 131)      73229       ['add_8[0][0]']                  \n",
      "                                                                                                  \n",
      " batch_normalization_48 (BatchN  (None, 30, 131)     524         ['conv1d_58[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_24 (ReLU)                (None, 30, 131)      0           ['batch_normalization_48[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_59 (Conv1D)             (None, 30, 131)      103097      ['re_lu_24[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_49 (BatchN  (None, 30, 131)     524         ['conv1d_59[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_25 (ReLU)                (None, 30, 131)      0           ['batch_normalization_49[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_60 (Conv1D)             (None, 30, 131)      103097      ['re_lu_25[0][0]']               \n",
      "                                                                                                  \n",
      " batch_normalization_50 (BatchN  (None, 30, 131)     524         ['conv1d_60[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " re_lu_26 (ReLU)                (None, 30, 131)      0           ['batch_normalization_50[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_61 (Conv1D)             (None, 30, 131)      17292       ['re_lu_26[0][0]']               \n",
      "                                                                                                  \n",
      " add_9 (Add)                    (None, 30, 131)      0           ['re_lu_26[0][0]',               \n",
      "                                                                  'conv1d_61[0][0]']              \n",
      "                                                                                                  \n",
      " global_average_pooling1d_2 (Gl  (None, 131)         0           ['add_9[0][0]']                  \n",
      " obalAveragePooling1D)                                                                            \n",
      "                                                                                                  \n",
      " dense_5 (Dense)                (None, 4)            528         ['global_average_pooling1d_2[0][0\n",
      "                                                                 ]']                              \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 726,505\n",
      "Trainable params: 724,471\n",
      "Non-trainable params: 2,034\n",
      "__________________________________________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 5 \n",
      "\n",
      "Model type: InceptionTime \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.00041757142225341446, 'regularization_rate': 0.0010252142188453691, 'network_depth': 5, 'filters_number': 89, 'max_kernel_size': 19}\n",
      " \n",
      "Model description:\n",
      "Model: \"model_3\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                   Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      " input_4 (InputLayer)           [(None, 30, 6)]      0           []                               \n",
      "                                                                                                  \n",
      " batch_normalization_51 (BatchN  (None, 30, 6)       24          ['input_4[0][0]']                \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " conv1d_62 (Conv1D)             (None, 30, 32)       192         ['batch_normalization_51[0][0]'] \n",
      "                                                                                                  \n",
      " max_pooling1d_3 (MaxPooling1D)  (None, 30, 6)       0           ['batch_normalization_51[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_63 (Conv1D)             (None, 30, 89)       54112       ['conv1d_62[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_64 (Conv1D)             (None, 30, 89)       25632       ['conv1d_62[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_65 (Conv1D)             (None, 30, 89)       11392       ['conv1d_62[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_66 (Conv1D)             (None, 30, 89)       534         ['max_pooling1d_3[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_3 (Concatenate)    (None, 30, 356)      0           ['conv1d_63[0][0]',              \n",
      "                                                                  'conv1d_64[0][0]',              \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                  'conv1d_65[0][0]',              \n",
      "                                                                  'conv1d_66[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_52 (BatchN  (None, 30, 356)     1424        ['concatenate_3[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_21 (Activation)     (None, 30, 356)      0           ['batch_normalization_52[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_67 (Conv1D)             (None, 30, 32)       11392       ['activation_21[0][0]']          \n",
      "                                                                                                  \n",
      " max_pooling1d_4 (MaxPooling1D)  (None, 30, 356)     0           ['activation_21[0][0]']          \n",
      "                                                                                                  \n",
      " conv1d_68 (Conv1D)             (None, 30, 89)       54112       ['conv1d_67[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_69 (Conv1D)             (None, 30, 89)       25632       ['conv1d_67[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_70 (Conv1D)             (None, 30, 89)       11392       ['conv1d_67[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_71 (Conv1D)             (None, 30, 89)       31684       ['max_pooling1d_4[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_4 (Concatenate)    (None, 30, 356)      0           ['conv1d_68[0][0]',              \n",
      "                                                                  'conv1d_69[0][0]',              \n",
      "                                                                  'conv1d_70[0][0]',              \n",
      "                                                                  'conv1d_71[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_53 (BatchN  (None, 30, 356)     1424        ['concatenate_4[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_22 (Activation)     (None, 30, 356)      0           ['batch_normalization_53[0][0]'] \n",
      "                                                                                                  \n",
      " conv1d_72 (Conv1D)             (None, 30, 32)       11392       ['activation_22[0][0]']          \n",
      "                                                                                                  \n",
      " max_pooling1d_5 (MaxPooling1D)  (None, 30, 356)     0           ['activation_22[0][0]']          \n",
      "                                                                                                  \n",
      " conv1d_73 (Conv1D)             (None, 30, 89)       54112       ['conv1d_72[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_74 (Conv1D)             (None, 30, 89)       25632       ['conv1d_72[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_75 (Conv1D)             (None, 30, 89)       11392       ['conv1d_72[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_76 (Conv1D)             (None, 30, 89)       31684       ['max_pooling1d_5[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_5 (Concatenate)    (None, 30, 356)      0           ['conv1d_73[0][0]',              \n",
      "                                                                  'conv1d_74[0][0]',              \n",
      "                                                                  'conv1d_75[0][0]',              \n",
      "                                                                  'conv1d_76[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_77 (Conv1D)             (None, 30, 356)      2136        ['batch_normalization_51[0][0]'] \n",
      "                                                                                                  \n",
      " batch_normalization_54 (BatchN  (None, 30, 356)     1424        ['concatenate_5[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " batch_normalization_55 (BatchN  (None, 30, 356)     1424        ['conv1d_77[0][0]']              \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_23 (Activation)     (None, 30, 356)      0           ['batch_normalization_54[0][0]'] \n",
      "                                                                                                  \n",
      " add_10 (Add)                   (None, 30, 356)      0           ['batch_normalization_55[0][0]', \n",
      "                                                                  'activation_23[0][0]']          \n",
      "                                                                                                  \n",
      " activation_24 (Activation)     (None, 30, 356)      0           ['add_10[0][0]']                 \n",
      "                                                                                                  \n",
      " conv1d_78 (Conv1D)             (None, 30, 32)       11392       ['activation_24[0][0]']          \n",
      "                                                                                                  \n",
      " max_pooling1d_6 (MaxPooling1D)  (None, 30, 356)     0           ['activation_24[0][0]']          \n",
      "                                                                                                  \n",
      " conv1d_79 (Conv1D)             (None, 30, 89)       54112       ['conv1d_78[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_80 (Conv1D)             (None, 30, 89)       25632       ['conv1d_78[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_81 (Conv1D)             (None, 30, 89)       11392       ['conv1d_78[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_82 (Conv1D)             (None, 30, 89)       31684       ['max_pooling1d_6[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_6 (Concatenate)    (None, 30, 356)      0           ['conv1d_79[0][0]',              \n",
      "                                                                  'conv1d_80[0][0]',              \n",
      "                                                                  'conv1d_81[0][0]',              \n",
      "                                                                  'conv1d_82[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_56 (BatchN  (None, 30, 356)     1424        ['concatenate_6[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_25 (Activation)     (None, 30, 356)      0           ['batch_normalization_56[0][0]'] \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                                                  \n",
      " conv1d_83 (Conv1D)             (None, 30, 32)       11392       ['activation_25[0][0]']          \n",
      "                                                                                                  \n",
      " max_pooling1d_7 (MaxPooling1D)  (None, 30, 356)     0           ['activation_25[0][0]']          \n",
      "                                                                                                  \n",
      " conv1d_84 (Conv1D)             (None, 30, 89)       54112       ['conv1d_83[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_85 (Conv1D)             (None, 30, 89)       25632       ['conv1d_83[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_86 (Conv1D)             (None, 30, 89)       11392       ['conv1d_83[0][0]']              \n",
      "                                                                                                  \n",
      " conv1d_87 (Conv1D)             (None, 30, 89)       31684       ['max_pooling1d_7[0][0]']        \n",
      "                                                                                                  \n",
      " concatenate_7 (Concatenate)    (None, 30, 356)      0           ['conv1d_84[0][0]',              \n",
      "                                                                  'conv1d_85[0][0]',              \n",
      "                                                                  'conv1d_86[0][0]',              \n",
      "                                                                  'conv1d_87[0][0]']              \n",
      "                                                                                                  \n",
      " batch_normalization_57 (BatchN  (None, 30, 356)     1424        ['concatenate_7[0][0]']          \n",
      " ormalization)                                                                                    \n",
      "                                                                                                  \n",
      " activation_26 (Activation)     (None, 30, 356)      0           ['batch_normalization_57[0][0]'] \n",
      "                                                                                                  \n",
      " global_average_pooling1d_3 (Gl  (None, 356)         0           ['activation_26[0][0]']          \n",
      " obalAveragePooling1D)                                                                            \n",
      "                                                                                                  \n",
      " dense_6 (Dense)                (None, 4)            1428        ['global_average_pooling1d_3[0][0\n",
      "                                                                 ]']                              \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 640,842\n",
      "Trainable params: 636,558\n",
      "Non-trainable params: 4,284\n",
      "__________________________________________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 6 \n",
      "\n",
      "Model type: CNN \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.0007902107985188801, 'regularization_rate': 0.00012774583804330792, 'filters': array([53, 42, 36, 18]), 'fc_hidden_nodes': 1622}\n",
      " \n",
      "Model description:\n",
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " batch_normalization_58 (Bat  (None, 30, 6)            24        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " conv1d_88 (Conv1D)          (None, 30, 53)            1007      \n",
      "                                                                 \n",
      " batch_normalization_59 (Bat  (None, 30, 53)           212       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_27 (Activation)  (None, 30, 53)            0         \n",
      "                                                                 \n",
      " conv1d_89 (Conv1D)          (None, 30, 42)            6720      \n",
      "                                                                 \n",
      " batch_normalization_60 (Bat  (None, 30, 42)           168       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_28 (Activation)  (None, 30, 42)            0         \n",
      "                                                                 \n",
      " conv1d_90 (Conv1D)          (None, 30, 36)            4572      \n",
      "                                                                 \n",
      " batch_normalization_61 (Bat  (None, 30, 36)           144       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_29 (Activation)  (None, 30, 36)            0         \n",
      "                                                                 \n",
      " conv1d_91 (Conv1D)          (None, 30, 18)            1962      \n",
      "                                                                 \n",
      " batch_normalization_62 (Bat  (None, 30, 18)           72        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_30 (Activation)  (None, 30, 18)            0         \n",
      "                                                                 \n",
      " flatten_1 (Flatten)         (None, 540)               0         \n",
      "                                                                 \n",
      " dense_7 (Dense)             (None, 1622)              877502    \n",
      "                                                                 \n",
      " activation_31 (Activation)  (None, 1622)              0         \n",
      "                                                                 \n",
      " dense_8 (Dense)             (None, 4)                 6492      \n",
      "                                                                 \n",
      " batch_normalization_63 (Bat  (None, 4)                16        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_32 (Activation)  (None, 4)                 0         \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 898,891\n",
      "Trainable params: 898,573\n",
      "Non-trainable params: 318\n",
      "_________________________________________________________________\n",
      " \n",
      "-------------------------------------------------------------------------------------------------------\n",
      "Model 7 \n",
      "\n",
      "Model type: DeepConvLSTM \n",
      "\n",
      "Hyperparameters:\n",
      "{'learning_rate': 0.005216294589208116, 'regularization_rate': 0.004895062158334504, 'filters': [80, 96, 80, 29, 66, 92, 11, 78, 50, 91], 'lstm_dims': [28, 94]}\n",
      " \n",
      "Model description:\n",
      "Model: \"sequential_3\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " batch_normalization_64 (Bat  (None, 30, 6)            24        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " reshape_2 (Reshape)         (None, 30, 6, 1)          0         \n",
      "                                                                 \n",
      " conv2d_4 (Conv2D)           (None, 30, 6, 80)         320       \n",
      "                                                                 \n",
      " batch_normalization_65 (Bat  (None, 30, 6, 80)        320       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_33 (Activation)  (None, 30, 6, 80)         0         \n",
      "                                                                 \n",
      " conv2d_5 (Conv2D)           (None, 30, 6, 96)         23136     \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                 \n",
      " batch_normalization_66 (Bat  (None, 30, 6, 96)        384       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_34 (Activation)  (None, 30, 6, 96)         0         \n",
      "                                                                 \n",
      " conv2d_6 (Conv2D)           (None, 30, 6, 80)         23120     \n",
      "                                                                 \n",
      " batch_normalization_67 (Bat  (None, 30, 6, 80)        320       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_35 (Activation)  (None, 30, 6, 80)         0         \n",
      "                                                                 \n",
      " conv2d_7 (Conv2D)           (None, 30, 6, 29)         6989      \n",
      "                                                                 \n",
      " batch_normalization_68 (Bat  (None, 30, 6, 29)        116       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_36 (Activation)  (None, 30, 6, 29)         0         \n",
      "                                                                 \n",
      " conv2d_8 (Conv2D)           (None, 30, 6, 66)         5808      \n",
      "                                                                 \n",
      " batch_normalization_69 (Bat  (None, 30, 6, 66)        264       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_37 (Activation)  (None, 30, 6, 66)         0         \n",
      "                                                                 \n",
      " conv2d_9 (Conv2D)           (None, 30, 6, 92)         18308     \n",
      "                                                                 \n",
      " batch_normalization_70 (Bat  (None, 30, 6, 92)        368       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_38 (Activation)  (None, 30, 6, 92)         0         \n",
      "                                                                 \n",
      " conv2d_10 (Conv2D)          (None, 30, 6, 11)         3047      \n",
      "                                                                 \n",
      " batch_normalization_71 (Bat  (None, 30, 6, 11)        44        \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_39 (Activation)  (None, 30, 6, 11)         0         \n",
      "                                                                 \n",
      " conv2d_11 (Conv2D)          (None, 30, 6, 78)         2652      \n",
      "                                                                 \n",
      " batch_normalization_72 (Bat  (None, 30, 6, 78)        312       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_40 (Activation)  (None, 30, 6, 78)         0         \n",
      "                                                                 \n",
      " conv2d_12 (Conv2D)          (None, 30, 6, 50)         11750     \n",
      "                                                                 \n",
      " batch_normalization_73 (Bat  (None, 30, 6, 50)        200       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_41 (Activation)  (None, 30, 6, 50)         0         \n",
      "                                                                 \n",
      " conv2d_13 (Conv2D)          (None, 30, 6, 91)         13741     \n",
      "                                                                 \n",
      " batch_normalization_74 (Bat  (None, 30, 6, 91)        364       \n",
      " chNormalization)                                                \n",
      "                                                                 \n",
      " activation_42 (Activation)  (None, 30, 6, 91)         0         \n",
      "                                                                 \n",
      " reshape_3 (Reshape)         (None, 30, 546)           0         \n",
      "                                                                 \n",
      " lstm_1 (LSTM)               (None, 30, 28)            64400     \n",
      "                                                                 \n",
      " lstm_2 (LSTM)               (None, 30, 94)            46248     \n",
      "                                                                 \n",
      " dropout_1 (Dropout)         (None, 30, 94)            0         \n",
      "                                                                 \n",
      " time_distributed_1 (TimeDis  (None, 30, 4)            380       \n",
      " tributed)                                                       \n",
      "                                                                 \n",
      " activation_43 (Activation)  (None, 30, 4)             0         \n",
      "                                                                 \n",
      " lambda_1 (Lambda)           (None, 4)                 0         \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 222,615\n",
      "Trainable params: 221,257\n",
      "Non-trainable params: 1,358\n",
      "_________________________________________________________________\n",
      " \n"
     ]
    }
   ],
   "source": [
    "models_to_print = range(len(models))\n",
    "for i, item in enumerate(models):\n",
    "    if i in models_to_print:\n",
    "        model, params, model_types = item\n",
    "        print(\"-------------------------------------------------------------------------------------------------------\")\n",
    "        print(\"Model \" + str(i), '\\n')          \n",
    "        print(\"Model type:\", model_types, '\\n')        \n",
    "        print(\"Hyperparameters:\")\n",
    "        print(params)\n",
    "        print(\" \")\n",
    "        print(\"Model description:\")\n",
    "        model.summary()        \n",
    "        print(\" \")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For a very quick view we can also just look the list of created models itself:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(<keras.engine.functional.Functional at 0x27264fc50c0>,\n",
       "  {'learning_rate': 0.005481464347736465,\n",
       "   'regularization_rate': 0.010209289921137055,\n",
       "   'network_depth': 5,\n",
       "   'min_filters_number': 107,\n",
       "   'max_kernel_size': 29},\n",
       "  'ResNet'),\n",
       " (<keras.engine.functional.Functional at 0x27265e7b5b0>,\n",
       "  {'learning_rate': 0.0436664988829779,\n",
       "   'regularization_rate': 0.003503259200059715,\n",
       "   'network_depth': 3,\n",
       "   'filters_number': 69,\n",
       "   'max_kernel_size': 17},\n",
       "  'InceptionTime'),\n",
       " (<keras.engine.sequential.Sequential at 0x2726529d4b0>,\n",
       "  {'learning_rate': 0.0394693672382836,\n",
       "   'regularization_rate': 0.002879158913712458,\n",
       "   'filters': [48, 43, 68, 77],\n",
       "   'lstm_dims': [78]},\n",
       "  'DeepConvLSTM'),\n",
       " (<keras.engine.sequential.Sequential at 0x27265e79fc0>,\n",
       "  {'learning_rate': 0.012851236876927944,\n",
       "   'regularization_rate': 0.013689688066052806,\n",
       "   'filters': array([ 41,  76,  90,  62,  86,  60,  14, 100,  73,  89]),\n",
       "   'fc_hidden_nodes': 1595},\n",
       "  'CNN'),\n",
       " (<keras.engine.functional.Functional at 0x27265445e70>,\n",
       "  {'learning_rate': 0.008923210639822017,\n",
       "   'regularization_rate': 0.003950367924123254,\n",
       "   'network_depth': 4,\n",
       "   'min_filters_number': 47,\n",
       "   'max_kernel_size': 17},\n",
       "  'ResNet'),\n",
       " (<keras.engine.functional.Functional at 0x272654613f0>,\n",
       "  {'learning_rate': 0.00041757142225341446,\n",
       "   'regularization_rate': 0.0010252142188453691,\n",
       "   'network_depth': 5,\n",
       "   'filters_number': 89,\n",
       "   'max_kernel_size': 19},\n",
       "  'InceptionTime'),\n",
       " (<keras.engine.sequential.Sequential at 0x27265447b50>,\n",
       "  {'learning_rate': 0.0007902107985188801,\n",
       "   'regularization_rate': 0.00012774583804330792,\n",
       "   'filters': array([53, 42, 36, 18]),\n",
       "   'fc_hidden_nodes': 1622},\n",
       "  'CNN'),\n",
       " (<keras.engine.sequential.Sequential at 0x27265461cc0>,\n",
       "  {'learning_rate': 0.005216294589208116,\n",
       "   'regularization_rate': 0.004895062158334504,\n",
       "   'filters': [80, 96, 80, 29, 66, 92, 11, 78, 50, 91],\n",
       "   'lstm_dims': [28, 94]},\n",
       "  'DeepConvLSTM')]"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *Question 3: Can you guess what hyperparameter 'learning rate' stands for?*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare models\n",
    "Now that the model architectures have been generated it is time to compare the models by training them on a subset of the training data and evaluating the models on the validation subset. This will help us to choose the best candidate model. The performance results for the models are stored in a json file, which we will visually inspect later on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define directory where the results, e.g. json file, will be stored\n",
    "resultpath = os.path.join(directory_to_extract_to, 'models')\n",
    "if not os.path.exists(resultpath):\n",
    "    os.makedirs(resultpath)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train models\n",
    "We are now going to train each of the models that we generated. On the one hand we want to train them as quickly as possible in order to be able to pick the best one as soon as possible. On the other hand we have to train each model long enough to get a good impression of its potential. As default mcfly uses **early stopping** which means that it will stop the training once the models accuracy on the validation set decreases. To better account for usually occuring fluctuations in the accuracy, we here set **early_stopping_patience = 5** which means that the model will stop if the validation accuracy has not improved since 5 epochs.\n",
    "\n",
    "We can influence the train time by adjusting the number of data samples that are used. This can be set with the argument 'subset_size'. We can also adjust the number of times the subset is iterated over. This number is called an epoch. For more complex dataset we recommend to start with no more than 5 epochs and a maximum subset size of 300. You can experiment with these numbers.\n",
    "\n",
    "Here, however, we work with a very small dataset (in terms of time resolution, channels, and number of samples). So we can run at longer training already.\n",
    "\n",
    "### The following training will take several minutes to run.\n",
    "This depends on the model architectures that were generated, the data used, and -obviously- the hardware this code is run on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated models will be trained on subset of the data (subset size: 300).\n",
      "Training model 0 ResNet\n"
     ]
    },
    {
     "ename": "ValueError",
     "evalue": "Invalid metric: \"['accuracy']\" is not among the metrics the models was compiled with ([]).",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "Cell \u001b[1;32mIn[42], line 4\u001b[0m\n\u001b[0;32m      1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01mmcfly\u001b[39;00m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01mfind_architecture\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m train_models_on_samples\n\u001b[0;32m      3\u001b[0m outputfile \u001b[38;5;241m=\u001b[39m os\u001b[38;5;241m.\u001b[39mpath\u001b[38;5;241m.\u001b[39mjoin(resultpath, \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mmodelcomparison.json\u001b[39m\u001b[38;5;124m'\u001b[39m)\n\u001b[1;32m----> 4\u001b[0m histories, val_accuracies, val_losses \u001b[38;5;241m=\u001b[39m \u001b[43mtrain_models_on_samples\u001b[49m\u001b[43m(\u001b[49m\u001b[43mX_train\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my_train_binary\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      5\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43mX_val\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my_val_binary\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m      6\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43mmodels\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mnr_epochs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m20\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[0;32m      7\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43msubset_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m300\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[0;32m      8\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43mearly_stopping_patience\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m \u001b[49m\u001b[38;5;241;43m5\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[0;32m      9\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43mverbose\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[0;32m     10\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43moutputfile\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moutputfile\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m     11\u001b[0m \u001b[43m                                                                \u001b[49m\u001b[43mmetric\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m[\u001b[49m\u001b[43mmetric\u001b[49m\u001b[43m]\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m     12\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mDetails of the training process were stored in \u001b[39m\u001b[38;5;124m'\u001b[39m,outputfile)\n",
      "File \u001b[1;32mc:\\users\\christiaanmeijer\\documents\\mcfly\\mcfly\\find_architecture.py:116\u001b[0m, in \u001b[0;36mtrain_models_on_samples\u001b[1;34m(X_train, y_train, X_val, y_val, models, nr_epochs, subset_size, verbose, outputfile, model_path, early_stopping_patience, batch_size, metric, class_weight)\u001b[0m\n\u001b[0;32m    114\u001b[0m model_metrics \u001b[38;5;241m=\u001b[39m [_get_metric_name(metric) \u001b[38;5;28;01mfor\u001b[39;00m metric \u001b[38;5;129;01min\u001b[39;00m model\u001b[38;5;241m.\u001b[39mmetrics_names]\n\u001b[0;32m    115\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m metric_name \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;129;01min\u001b[39;00m model_metrics:\n\u001b[1;32m--> 116\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mInvalid metric: \u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m is not among the metrics the models was compiled with (\u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[38;5;124m).\u001b[39m\u001b[38;5;124m'\u001b[39m\n\u001b[0;32m    117\u001b[0m                      \u001b[38;5;241m.\u001b[39mformat(metric_name, model_metrics))\n\u001b[0;32m    118\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m early_stopping_patience \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[0;32m    119\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m early_stopping_patience \u001b[38;5;241m==\u001b[39m \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mauto\u001b[39m\u001b[38;5;124m'\u001b[39m:\n",
      "\u001b[1;31mValueError\u001b[0m: Invalid metric: \"['accuracy']\" is not among the metrics the models was compiled with ([])."
     ]
    }
   ],
   "source": [
    "from mcfly.find_architecture import train_models_on_samples\n",
    "\n",
    "outputfile = os.path.join(resultpath, 'modelcomparison.json')\n",
    "histories, val_accuracies, val_losses = train_models_on_samples(X_train, y_train_binary,\n",
    "                                                                X_val, y_val_binary,\n",
    "                                                                models, nr_epochs=20,\n",
    "                                                                subset_size=300,\n",
    "                                                                early_stopping_patience = 5,\n",
    "                                                                verbose=True,\n",
    "                                                                outputfile=outputfile,\n",
    "                                                                metric=[metric])\n",
    "print('Details of the training process were stored in ',outputfile)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models[0][0].metrics_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "models[0][0].compile?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *Question 4: What do the terms 'acc' and 'loss' in the output refer to?*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inspect model performance (Visualization)\n",
    "\n",
    "We can inspect the learning process in the visualization tool on http://nlesc.github.io/mcfly/.\n",
    "\n",
    "Alternatively, you can run the visualization from a local web service:\n",
    "- Clone the mcfly github repository (if you haven't done so already for visualization)\n",
    "\n",
    " `git clone https://github.com/NLeSC/mcfly`\n",
    "\n",
    "\n",
    "- navigate to the html folder:\n",
    "\n",
    " `cd mcfly/html`\n",
    "\n",
    "\n",
    "- Start a web server. This can be done in various ways, for example:\n",
    "`python3 -m http.server`\n",
    "\n",
    "Notice the port number the web server is serving on. This is usually 8000.\n",
    "With a web browser, navigate to [localhost:8000](localhost:8000). \n",
    "\n",
    "You need to upload the json file that contains the details of the training process. The following line of code shows the path to this file:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'.\\\\data\\\\models\\\\modelcomparison.json'"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "outputfile"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *Question 5:  Look at the visualization. Which model performs best?*\n",
    "### *Question 6:  Did you train all models with a sufficient number of iterations?*\n",
    "### *Question 7:  Are there cases of clear overfitting?*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inspect model performance (table)\n",
    "\n",
    "Let's compare the performance of the models by showing the results as a table."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "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>model</th>\n",
       "      <th>model-type</th>\n",
       "      <th>train_accuracy</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>val_accuracy</th>\n",
       "      <th>val_loss</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>{'learning_rate': 0.005481464347736465, 'regul...</td>\n",
       "      <td>ResNet</td>\n",
       "      <td>0.450331</td>\n",
       "      <td>1.090753</td>\n",
       "      <td>0.253333</td>\n",
       "      <td>10.486857</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>{'learning_rate': 0.0436664988829779, 'regular...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>0.980132</td>\n",
       "      <td>0.069513</td>\n",
       "      <td>0.360000</td>\n",
       "      <td>30.067336</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>{'learning_rate': 0.0394693672382836, 'regular...</td>\n",
       "      <td>DeepConvLSTM</td>\n",
       "      <td>0.205298</td>\n",
       "      <td>1.621776</td>\n",
       "      <td>0.226667</td>\n",
       "      <td>1.492279</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>{'learning_rate': 0.012851236876927944, 'regul...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>0.841060</td>\n",
       "      <td>1.361104</td>\n",
       "      <td>0.226667</td>\n",
       "      <td>3.304446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>{'learning_rate': 0.008923210639822017, 'regul...</td>\n",
       "      <td>ResNet</td>\n",
       "      <td>0.913907</td>\n",
       "      <td>0.207766</td>\n",
       "      <td>0.653333</td>\n",
       "      <td>224.207554</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>{'learning_rate': 0.00041757142225341446, 'reg...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.005901</td>\n",
       "      <td>0.906667</td>\n",
       "      <td>0.347267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>{'learning_rate': 0.0007902107985188801, 'regu...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>0.993378</td>\n",
       "      <td>0.478264</td>\n",
       "      <td>0.760000</td>\n",
       "      <td>1.037439</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>{'learning_rate': 0.005216294589208116, 'regul...</td>\n",
       "      <td>DeepConvLSTM</td>\n",
       "      <td>0.827815</td>\n",
       "      <td>0.758923</td>\n",
       "      <td>0.640000</td>\n",
       "      <td>1.303036</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                               model     model-type  \\\n",
       "0  {'learning_rate': 0.005481464347736465, 'regul...         ResNet   \n",
       "1  {'learning_rate': 0.0436664988829779, 'regular...  InceptionTime   \n",
       "2  {'learning_rate': 0.0394693672382836, 'regular...   DeepConvLSTM   \n",
       "3  {'learning_rate': 0.012851236876927944, 'regul...            CNN   \n",
       "4  {'learning_rate': 0.008923210639822017, 'regul...         ResNet   \n",
       "5  {'learning_rate': 0.00041757142225341446, 'reg...  InceptionTime   \n",
       "6  {'learning_rate': 0.0007902107985188801, 'regu...            CNN   \n",
       "7  {'learning_rate': 0.005216294589208116, 'regul...   DeepConvLSTM   \n",
       "\n",
       "   train_accuracy  train_loss  val_accuracy    val_loss  \n",
       "0        0.450331    1.090753      0.253333   10.486857  \n",
       "1        0.980132    0.069513      0.360000   30.067336  \n",
       "2        0.205298    1.621776      0.226667    1.492279  \n",
       "3        0.841060    1.361104      0.226667    3.304446  \n",
       "4        0.913907    0.207766      0.653333  224.207554  \n",
       "5        1.000000    0.005901      0.906667    0.347267  \n",
       "6        0.993378    0.478264      0.760000    1.037439  \n",
       "7        0.827815    0.758923      0.640000    1.303036  "
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metric = 'accuracy'\n",
    "modelcomparisons = pd.DataFrame({'model':[str(params) for model, params, model_types in models],\n",
    "                                'model-type':[str(model_types) for model, params, model_types in models],\n",
    "                                'train_{}'.format(metric): [history.history[metric][-1] for history in histories],\n",
    "                                'train_loss': [history.history['loss'][-1] for history in histories],\n",
    "                                'val_{}'.format(metric): [history.history['val_{}'.format(metric)][-1] for history in histories],\n",
    "                                'val_loss': [history.history['val_loss'][-1] for history in histories]\n",
    "                                })\n",
    "modelcomparisons.to_csv(os.path.join(resultpath, 'modelcomparisons.csv'))\n",
    "\n",
    "modelcomparisons"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Next steps:\n",
    "## 1) Continue iteratively: generate more models of desired type\n",
    "*mcfly* is generating models of 4 different architecture types (CNN, DeepConvLSTM, ResNet, InceptionTime). There performance of course strongly depends on the hyperparameters used, but they also all have different strengths and weaknesses. Often it makes sense to select the most promising type(s) after the first screening we did above and continue to explore options further.\n",
    "\n",
    "### Narrow the search: model types\n",
    "In this case, let's say that the most promising models were of typ CNN and InceptionTime. So we will ignore the other architecture for now.\n",
    "We can do this in *mcfly* by specifing which models we want to generate: ``model_types = ['CNN', 'InceptionTime']``\n",
    "\n",
    "### Narrow the search: hyperparameters\n",
    "Figuring out which hyperparameters work better than others can be very complicated since often there exist complex links between them. But let's say we already figured something out, e.g. that high learning rates lead to overfitting on our data. Then *mcfly* easily allows to restrict the parameter search space!\n",
    "Simply add the new desired range to the generate_models function: ``low_lr=2``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The value of model_types is set from ['CNN', 'DeepConvLSTM', 'ResNet', 'InceptionTime'] (default) to ['CNN', 'InceptionTime']\n",
      "The value of low_lr is set from 1 (default) to 2\n",
      "The value of metrics is set from ['accuracy'] (default) to ['accuracy']\n",
      "Set maximum kernel size for InceptionTime models to number of timesteps.\n"
     ]
    }
   ],
   "source": [
    "num_classes = y_train_binary.shape[1]\n",
    "metric = 'accuracy'\n",
    "models = mcfly.modelgen.generate_models(X_train.shape,\n",
    "                                        number_of_classes=num_classes,\n",
    "                                        number_of_models = 8,\n",
    "                                        model_types = ['CNN', 'InceptionTime'],\n",
    "                                        low_lr=2,\n",
    "                                        metrics=[metric])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We again now have to train those newly generated models:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated models will be trained on subset of the data (subset size: 300).\n",
      "Training model 0 InceptionTime\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 7s 44ms/sample - loss: 1.1319 - accuracy: 0.5762 - val_loss: 2.0306 - val_accuracy: 0.2667\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.6355 - accuracy: 0.8609 - val_loss: 2.6487 - val_accuracy: 0.2667\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.4208 - accuracy: 0.9007 - val_loss: 1.8872 - val_accuracy: 0.3067\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.2631 - accuracy: 0.9470 - val_loss: 1.5133 - val_accuracy: 0.4133\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.1977 - accuracy: 0.9735 - val_loss: 1.1984 - val_accuracy: 0.6267\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.1306 - accuracy: 0.9868 - val_loss: 1.2518 - val_accuracy: 0.6267\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0877 - accuracy: 0.9934 - val_loss: 0.8640 - val_accuracy: 0.6800\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0502 - accuracy: 1.0000 - val_loss: 0.8586 - val_accuracy: 0.6800\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0334 - accuracy: 1.0000 - val_loss: 0.8258 - val_accuracy: 0.7067\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.0431 - accuracy: 1.0000 - val_loss: 0.6251 - val_accuracy: 0.7600\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0271 - accuracy: 1.0000 - val_loss: 0.4829 - val_accuracy: 0.8667\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0279 - accuracy: 1.0000 - val_loss: 0.4965 - val_accuracy: 0.8267\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0128 - accuracy: 1.0000 - val_loss: 0.4089 - val_accuracy: 0.8800\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.0186 - accuracy: 1.0000 - val_loss: 0.5136 - val_accuracy: 0.8267\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.0106 - accuracy: 1.0000 - val_loss: 0.4854 - val_accuracy: 0.8400\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0090 - accuracy: 1.0000 - val_loss: 0.3895 - val_accuracy: 0.8667\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0156 - accuracy: 1.0000 - val_loss: 0.3852 - val_accuracy: 0.8667\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0077 - accuracy: 1.0000 - val_loss: 0.3925 - val_accuracy: 0.8800\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 1s 7ms/sample - loss: 0.0089 - accuracy: 1.0000 - val_loss: 0.3956 - val_accuracy: 0.8667\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 1s 6ms/sample - loss: 0.0056 - accuracy: 1.0000 - val_loss: 0.3809 - val_accuracy: 0.8667\n",
      "Training model 1 CNN\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 6s 36ms/sample - loss: 1.4506 - accuracy: 0.5033 - val_loss: 19.5277 - val_accuracy: 0.3600\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 1.0763 - accuracy: 0.8874 - val_loss: 9.6035 - val_accuracy: 0.3867\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 1.0279 - accuracy: 0.9404 - val_loss: 8.0702 - val_accuracy: 0.3867\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 1s 3ms/sample - loss: 0.9356 - accuracy: 0.9868 - val_loss: 4.7908 - val_accuracy: 0.4133\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.8706 - accuracy: 0.9934 - val_loss: 4.1096 - val_accuracy: 0.3600\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.7594 - accuracy: 1.0000 - val_loss: 3.4770 - val_accuracy: 0.3333\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.7055 - accuracy: 1.0000 - val_loss: 2.3531 - val_accuracy: 0.4533\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 1s 5ms/sample - loss: 0.6364 - accuracy: 1.0000 - val_loss: 2.0703 - val_accuracy: 0.4533\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.6012 - accuracy: 1.0000 - val_loss: 1.6075 - val_accuracy: 0.5067\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 1s 3ms/sample - loss: 0.5318 - accuracy: 1.0000 - val_loss: 1.2024 - val_accuracy: 0.6800\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.5132 - accuracy: 1.0000 - val_loss: 1.0171 - val_accuracy: 0.7867\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.4608 - accuracy: 1.0000 - val_loss: 0.9910 - val_accuracy: 0.7333\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.4499 - accuracy: 1.0000 - val_loss: 0.8402 - val_accuracy: 0.7867\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 1s 4ms/sample - loss: 0.3943 - accuracy: 1.0000 - val_loss: 0.8020 - val_accuracy: 0.8533\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.3728 - accuracy: 1.0000 - val_loss: 0.7608 - val_accuracy: 0.8933\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 1s 3ms/sample - loss: 0.3502 - accuracy: 0.9934 - val_loss: 0.7521 - val_accuracy: 0.8667\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.3731 - accuracy: 0.9934 - val_loss: 0.7830 - val_accuracy: 0.7867\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.3640 - accuracy: 0.9868 - val_loss: 0.6484 - val_accuracy: 0.8667\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 1s 3ms/sample - loss: 0.4089 - accuracy: 0.9801 - val_loss: 0.7371 - val_accuracy: 0.8133\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.4028 - accuracy: 0.9735 - val_loss: 0.7121 - val_accuracy: 0.8800\n",
      "Training model 2 InceptionTime\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 11s 72ms/sample - loss: 0.8559 - accuracy: 0.6954 - val_loss: 10.4376 - val_accuracy: 0.3600\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 1s 9ms/sample - loss: 0.1593 - accuracy: 0.9536 - val_loss: 22.4276 - val_accuracy: 0.2667\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 1s 9ms/sample - loss: 0.1006 - accuracy: 0.9536 - val_loss: 25.8605 - val_accuracy: 0.2667\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 1s 9ms/sample - loss: 0.0261 - accuracy: 0.9934 - val_loss: 23.2571 - val_accuracy: 0.2667\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 1s 9ms/sample - loss: 0.0336 - accuracy: 0.9868 - val_loss: 17.9154 - val_accuracy: 0.2800\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 1s 9ms/sample - loss: 0.0194 - accuracy: 0.9934 - val_loss: 12.1619 - val_accuracy: 0.3467\n",
      "Epoch 00006: early stopping\n",
      "Training model 3 CNN\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 11.3297 - accuracy: 0.6026 - val_loss: 10.9818 - val_accuracy: 0.7067\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 0s 984us/sample - loss: 7.6320 - accuracy: 0.9669 - val_loss: 7.1768 - val_accuracy: 0.7333\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 5.1260 - accuracy: 0.9470 - val_loss: 4.9807 - val_accuracy: 0.7600\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 3.6269 - accuracy: 0.9669 - val_loss: 3.7145 - val_accuracy: 0.7333\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 2.6814 - accuracy: 0.9934 - val_loss: 2.7587 - val_accuracy: 0.7600\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 2.0734 - accuracy: 0.9868 - val_loss: 2.3245 - val_accuracy: 0.7733\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 1.6868 - accuracy: 0.9934 - val_loss: 2.0121 - val_accuracy: 0.7600\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 0s 2ms/sample - loss: 1.4953 - accuracy: 0.9801 - val_loss: 1.8292 - val_accuracy: 0.7733\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 9/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 1.3838 - accuracy: 0.9470 - val_loss: 1.7578 - val_accuracy: 0.7600\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 1.2920 - accuracy: 0.9404 - val_loss: 1.6947 - val_accuracy: 0.7333\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 1.1562 - accuracy: 0.9735 - val_loss: 1.5207 - val_accuracy: 0.7733\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 1.0927 - accuracy: 0.9735 - val_loss: 1.4505 - val_accuracy: 0.7733\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.9933 - accuracy: 0.9669 - val_loss: 1.4315 - val_accuracy: 0.7467\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.8713 - accuracy: 1.0000 - val_loss: 1.3224 - val_accuracy: 0.7467\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.8108 - accuracy: 0.9801 - val_loss: 1.3399 - val_accuracy: 0.7200\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.7695 - accuracy: 0.9669 - val_loss: 1.2536 - val_accuracy: 0.7733\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.7869 - accuracy: 0.9470 - val_loss: 1.3183 - val_accuracy: 0.7067\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.7210 - accuracy: 1.0000 - val_loss: 1.3130 - val_accuracy: 0.7333\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.6931 - accuracy: 0.9934 - val_loss: 1.2476 - val_accuracy: 0.7333\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.6010 - accuracy: 0.9868 - val_loss: 1.2888 - val_accuracy: 0.6267\n",
      "Training model 4 CNN\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 4s 24ms/sample - loss: 3.9158 - accuracy: 0.5828 - val_loss: 10.8585 - val_accuracy: 0.6400\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 5.0624 - accuracy: 0.9073 - val_loss: 8.0933 - val_accuracy: 0.6533\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 4.8143 - accuracy: 0.9735 - val_loss: 6.4836 - val_accuracy: 0.6533\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 4.0192 - accuracy: 0.9603 - val_loss: 4.9941 - val_accuracy: 0.6400\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 3.1182 - accuracy: 0.9934 - val_loss: 3.4103 - val_accuracy: 0.6800\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 2.4347 - accuracy: 0.9934 - val_loss: 2.8470 - val_accuracy: 0.6667\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 1.9041 - accuracy: 0.9934 - val_loss: 2.1540 - val_accuracy: 0.7467\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 1.5581 - accuracy: 0.9868 - val_loss: 1.8004 - val_accuracy: 0.7867\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 1.2535 - accuracy: 0.9868 - val_loss: 1.4886 - val_accuracy: 0.8400\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.9934 - accuracy: 1.0000 - val_loss: 1.4281 - val_accuracy: 0.8400\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.8260 - accuracy: 1.0000 - val_loss: 1.3586 - val_accuracy: 0.7600\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.6940 - accuracy: 1.0000 - val_loss: 1.3125 - val_accuracy: 0.7467\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.6078 - accuracy: 0.9934 - val_loss: 1.2961 - val_accuracy: 0.6933\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.5777 - accuracy: 0.9934 - val_loss: 1.2914 - val_accuracy: 0.7467\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.5300 - accuracy: 0.9934 - val_loss: 1.1098 - val_accuracy: 0.7867\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.5047 - accuracy: 0.9801 - val_loss: 0.9791 - val_accuracy: 0.8133\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.4539 - accuracy: 0.9868 - val_loss: 1.1643 - val_accuracy: 0.7467\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.4106 - accuracy: 0.9934 - val_loss: 1.2082 - val_accuracy: 0.5867\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.4151 - accuracy: 0.9735 - val_loss: 1.2566 - val_accuracy: 0.5600\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 0s 3ms/sample - loss: 0.4337 - accuracy: 0.9669 - val_loss: 1.0199 - val_accuracy: 0.7333\n",
      "Training model 5 InceptionTime\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 8s 56ms/sample - loss: 0.8051 - accuracy: 0.7152 - val_loss: 4.4876 - val_accuracy: 0.4667\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.2252 - accuracy: 0.9404 - val_loss: 4.0818 - val_accuracy: 0.4000\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1738 - accuracy: 0.9404 - val_loss: 3.8690 - val_accuracy: 0.4133\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0562 - accuracy: 1.0000 - val_loss: 3.4122 - val_accuracy: 0.3733\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.0461 - accuracy: 0.9934 - val_loss: 3.3737 - val_accuracy: 0.4800\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0226 - accuracy: 1.0000 - val_loss: 1.7626 - val_accuracy: 0.7067\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0209 - accuracy: 1.0000 - val_loss: 1.6895 - val_accuracy: 0.5600\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0172 - accuracy: 1.0000 - val_loss: 1.9631 - val_accuracy: 0.4400\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.0126 - accuracy: 1.0000 - val_loss: 1.8519 - val_accuracy: 0.6400\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0060 - accuracy: 1.0000 - val_loss: 0.6194 - val_accuracy: 0.7733\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.0080 - accuracy: 1.0000 - val_loss: 0.9216 - val_accuracy: 0.7600\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.0056 - accuracy: 1.0000 - val_loss: 1.0721 - val_accuracy: 0.7600\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.7616 - val_accuracy: 0.7600\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.5770 - val_accuracy: 0.8000\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0010 - accuracy: 1.0000 - val_loss: 0.5115 - val_accuracy: 0.8133\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 2s 14ms/sample - loss: 0.0041 - accuracy: 1.0000 - val_loss: 0.6063 - val_accuracy: 0.8133\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0280 - accuracy: 0.9868 - val_loss: 0.4081 - val_accuracy: 0.8800\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1736 - accuracy: 0.9669 - val_loss: 2.3490 - val_accuracy: 0.6267\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1185 - accuracy: 0.9536 - val_loss: 1.2047 - val_accuracy: 0.7200\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0389 - accuracy: 0.9868 - val_loss: 0.8267 - val_accuracy: 0.7867\n",
      "Training model 6 CNN\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 1.2833 - accuracy: 0.5563 - val_loss: 1.2578 - val_accuracy: 0.6667\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.7191 - accuracy: 0.9272 - val_loss: 1.2042 - val_accuracy: 0.6933\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.6525 - accuracy: 0.9470 - val_loss: 1.1114 - val_accuracy: 0.7333\n",
      "Epoch 4/20\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.5805 - accuracy: 0.9801 - val_loss: 1.0561 - val_accuracy: 0.7600\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.6016 - accuracy: 0.9735 - val_loss: 1.0097 - val_accuracy: 0.7600\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.5353 - accuracy: 1.0000 - val_loss: 0.9793 - val_accuracy: 0.7600\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.5235 - accuracy: 1.0000 - val_loss: 0.9448 - val_accuracy: 0.7600\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.5306 - accuracy: 1.0000 - val_loss: 0.9169 - val_accuracy: 0.7733\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.5152 - accuracy: 1.0000 - val_loss: 0.8912 - val_accuracy: 0.7867\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4850 - accuracy: 1.0000 - val_loss: 0.8709 - val_accuracy: 0.7867\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4665 - accuracy: 1.0000 - val_loss: 0.8511 - val_accuracy: 0.8000\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4963 - accuracy: 1.0000 - val_loss: 0.8370 - val_accuracy: 0.8000\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4689 - accuracy: 1.0000 - val_loss: 0.8386 - val_accuracy: 0.8133\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4885 - accuracy: 1.0000 - val_loss: 0.8381 - val_accuracy: 0.8133\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4641 - accuracy: 1.0000 - val_loss: 0.8270 - val_accuracy: 0.8000\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4667 - accuracy: 1.0000 - val_loss: 0.8176 - val_accuracy: 0.8133\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4548 - accuracy: 1.0000 - val_loss: 0.8171 - val_accuracy: 0.8000\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4743 - accuracy: 1.0000 - val_loss: 0.8184 - val_accuracy: 0.8000\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4606 - accuracy: 1.0000 - val_loss: 0.8171 - val_accuracy: 0.8000\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 0s 1ms/sample - loss: 0.4536 - accuracy: 1.0000 - val_loss: 0.8178 - val_accuracy: 0.8000\n",
      "Training model 7 InceptionTime\n",
      "Train on 151 samples, validate on 75 samples\n",
      "Epoch 1/20\n",
      "151/151 [==============================] - 11s 70ms/sample - loss: 0.9074 - accuracy: 0.6821 - val_loss: 415.6523 - val_accuracy: 0.2667\n",
      "Epoch 2/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.5056 - accuracy: 0.7815 - val_loss: 598.1390 - val_accuracy: 0.2667\n",
      "Epoch 3/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.2507 - accuracy: 0.9205 - val_loss: 420.4679 - val_accuracy: 0.2667\n",
      "Epoch 4/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1646 - accuracy: 0.9404 - val_loss: 191.3671 - val_accuracy: 0.2667\n",
      "Epoch 5/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1482 - accuracy: 0.9470 - val_loss: 94.2274 - val_accuracy: 0.2667\n",
      "Epoch 6/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0368 - accuracy: 0.9934 - val_loss: 128.1895 - val_accuracy: 0.2667\n",
      "Epoch 7/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0573 - accuracy: 0.9735 - val_loss: 105.8133 - val_accuracy: 0.2400\n",
      "Epoch 8/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 0.1431 - accuracy: 0.9272 - val_loss: 168.4964 - val_accuracy: 0.2667\n",
      "Epoch 9/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1917 - accuracy: 0.9139 - val_loss: 58.7867 - val_accuracy: 0.2667\n",
      "Epoch 10/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.2187 - accuracy: 0.9007 - val_loss: 81.1334 - val_accuracy: 0.2667\n",
      "Epoch 11/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.1943 - accuracy: 0.9603 - val_loss: 82.2359 - val_accuracy: 0.2933\n",
      "Epoch 12/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0645 - accuracy: 0.9868 - val_loss: 84.1961 - val_accuracy: 0.2267\n",
      "Epoch 13/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0260 - accuracy: 1.0000 - val_loss: 49.4664 - val_accuracy: 0.3467\n",
      "Epoch 14/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 0.0160 - accuracy: 1.0000 - val_loss: 26.9527 - val_accuracy: 0.4800\n",
      "Epoch 15/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 0.0081 - accuracy: 1.0000 - val_loss: 16.9202 - val_accuracy: 0.4933\n",
      "Epoch 16/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0042 - accuracy: 1.0000 - val_loss: 10.6787 - val_accuracy: 0.5600\n",
      "Epoch 17/20\n",
      "151/151 [==============================] - 3s 17ms/sample - loss: 0.0014 - accuracy: 1.0000 - val_loss: 7.1064 - val_accuracy: 0.6000\n",
      "Epoch 18/20\n",
      "151/151 [==============================] - 2s 16ms/sample - loss: 0.0035 - accuracy: 1.0000 - val_loss: 4.9386 - val_accuracy: 0.6400\n",
      "Epoch 19/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0015 - accuracy: 1.0000 - val_loss: 3.5662 - val_accuracy: 0.6933\n",
      "Epoch 20/20\n",
      "151/151 [==============================] - 2s 15ms/sample - loss: 0.0173 - accuracy: 0.9934 - val_loss: 1.7428 - val_accuracy: 0.7200\n"
     ]
    }
   ],
   "source": [
    "outputfile = os.path.join(resultpath, 'modelcomparison_2.json')\n",
    "histories, val_accuracies, val_losses = train_models_on_samples(X_train, y_train_binary,\n",
    "                                                                X_val, y_val_binary,\n",
    "                                                                models, nr_epochs=20,\n",
    "                                                                subset_size=300,\n",
    "                                                                early_stopping_patience = 5,\n",
    "                                                                verbose=True,\n",
    "                                                                outputfile=outputfile,\n",
    "                                                                metric=metric)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>model</th>\n",
       "      <th>model-type</th>\n",
       "      <th>train_accuracy</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>val_accuracy</th>\n",
       "      <th>val_loss</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>{'learning_rate': 0.0002508899689375268, 'regu...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.005574</td>\n",
       "      <td>0.866667</td>\n",
       "      <td>0.380934</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>{'learning_rate': 0.008102929671455617, 'regul...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>0.973510</td>\n",
       "      <td>0.402825</td>\n",
       "      <td>0.880000</td>\n",
       "      <td>0.712099</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>{'learning_rate': 0.0016297966098709998, 'regu...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>0.993378</td>\n",
       "      <td>0.019420</td>\n",
       "      <td>0.346667</td>\n",
       "      <td>12.161851</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>{'learning_rate': 0.0019430892193161443, 'regu...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>0.986755</td>\n",
       "      <td>0.601033</td>\n",
       "      <td>0.626667</td>\n",
       "      <td>1.288750</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>{'learning_rate': 0.008889925393993863, 'regul...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>0.966887</td>\n",
       "      <td>0.433666</td>\n",
       "      <td>0.733333</td>\n",
       "      <td>1.019946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>{'learning_rate': 0.0007907590834387294, 'regu...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>0.986755</td>\n",
       "      <td>0.038861</td>\n",
       "      <td>0.786667</td>\n",
       "      <td>0.826676</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>{'learning_rate': 0.0005081261465795578, 'regu...</td>\n",
       "      <td>CNN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.453622</td>\n",
       "      <td>0.800000</td>\n",
       "      <td>0.817783</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>{'learning_rate': 0.0045608494108193166, 'regu...</td>\n",
       "      <td>InceptionTime</td>\n",
       "      <td>0.993378</td>\n",
       "      <td>0.017331</td>\n",
       "      <td>0.720000</td>\n",
       "      <td>1.742814</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                               model     model-type  \\\n",
       "0  {'learning_rate': 0.0002508899689375268, 'regu...  InceptionTime   \n",
       "1  {'learning_rate': 0.008102929671455617, 'regul...            CNN   \n",
       "2  {'learning_rate': 0.0016297966098709998, 'regu...  InceptionTime   \n",
       "3  {'learning_rate': 0.0019430892193161443, 'regu...            CNN   \n",
       "4  {'learning_rate': 0.008889925393993863, 'regul...            CNN   \n",
       "5  {'learning_rate': 0.0007907590834387294, 'regu...  InceptionTime   \n",
       "6  {'learning_rate': 0.0005081261465795578, 'regu...            CNN   \n",
       "7  {'learning_rate': 0.0045608494108193166, 'regu...  InceptionTime   \n",
       "\n",
       "   train_accuracy  train_loss  val_accuracy   val_loss  \n",
       "0        1.000000    0.005574      0.866667   0.380934  \n",
       "1        0.973510    0.402825      0.880000   0.712099  \n",
       "2        0.993378    0.019420      0.346667  12.161851  \n",
       "3        0.986755    0.601033      0.626667   1.288750  \n",
       "4        0.966887    0.433666      0.733333   1.019946  \n",
       "5        0.986755    0.038861      0.786667   0.826676  \n",
       "6        1.000000    0.453622      0.800000   0.817783  \n",
       "7        0.993378    0.017331      0.720000   1.742814  "
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metric = 'accuracy'\n",
    "modelcomparisons_2 = pd.DataFrame({'model':[str(params) for model, params, model_types in models],\n",
    "                                'model-type':[str(model_types) for model, params, model_types in models],\n",
    "                                'train_{}'.format(metric): [history.history[metric][-1] for history in histories],\n",
    "                                'train_loss': [history.history['loss'][-1] for history in histories],\n",
    "                                'val_{}'.format(metric): [history.history['val_{}'.format(metric)][-1] for history in histories],\n",
    "                                'val_loss': [history.history['val_loss'][-1] for history in histories]\n",
    "                                })\n",
    "modelcomparisons_2.to_csv(os.path.join(resultpath, 'modelcomparisons_2.csv'))\n",
    "\n",
    "modelcomparisons_2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## 2) Chose the best model and train longer\n",
    "Now that we found an effective architecture, we can choose the most promising model. For example, we can choose the model with the highest accuracy on the validation data set. To maximize this models performance, we will train this model on more data and more epochs.\n",
    "\n",
    "### data subset --> full dataset\n",
    "If we did the search above on a subset of the data only, than now is the time to train on the full dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model type and parameters of the best model:\n",
      "CNN\n",
      "{'learning_rate': 0.008102929671455617, 'regularization_rate': 0.00011234135658673196, 'filters': array([69, 27, 25, 40, 87, 36]), 'fc_hidden_nodes': 1841}\n"
     ]
    }
   ],
   "source": [
    "best_model_index = np.argmax(val_accuracies)\n",
    "best_model, best_params, best_model_types = models[best_model_index]\n",
    "print('Model type and parameters of the best model:')\n",
    "print(best_model_types)\n",
    "print(best_params)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train the best model on the full dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now that we have identified the best model architecture out of our random pool of models we can continue by training the model on the full training set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(<tensorflow.python.keras.engine.sequential.Sequential at 0x1d9998fc688>,\n",
       " {'learning_rate': 0.008102929671455617,\n",
       "  'regularization_rate': 0.00011234135658673196,\n",
       "  'filters': array([69, 27, 25, 40, 87, 36]),\n",
       "  'fc_hidden_nodes': 1841},\n",
       " 'CNN')"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models[best_model_index]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training model 0 CNN\n",
      "Train on 150 samples, validate on 75 samples\n",
      "Epoch 1/50\n",
      "150/150 [==============================] - 0s 2ms/sample - loss: 0.2637 - accuracy: 0.9667 - val_loss: 0.7022 - val_accuracy: 0.8400\n",
      "Epoch 2/50\n",
      "150/150 [==============================] - 0s 3ms/sample - loss: 0.2750 - accuracy: 0.9867 - val_loss: 0.6932 - val_accuracy: 0.8400\n",
      "Epoch 3/50\n",
      "150/150 [==============================] - 0s 3ms/sample - loss: 0.3578 - accuracy: 0.9533 - val_loss: 0.8209 - val_accuracy: 0.7600\n",
      "Epoch 4/50\n",
      "150/150 [==============================] - 0s 3ms/sample - loss: 0.3415 - accuracy: 0.9667 - val_loss: 4.1832 - val_accuracy: 0.5200\n",
      "Epoch 00004: early stopping\n"
     ]
    }
   ],
   "source": [
    "#We make a copy of the model, to start training from fresh\n",
    "_,_,_ = train_models_on_samples(X_train, y_train_binary,\n",
    "                                X_val, y_val_binary,\n",
    "                                [models[best_model_index]], nr_epochs=50,\n",
    "                                subset_size=None,\n",
    "                                early_stopping_patience = 2,\n",
    "                                verbose=True,\n",
    "                                outputfile=outputfile,\n",
    "                                metric=metric)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *Question 7: Do you think it is useful to train with more than 1 epoch?*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving, loading and comparing reloaded model with original model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The model can be saved for future use. The savemodel function will save two separate files: a json file for the architecture and a npy (numpy array) file for the weights."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelname = 'my_bestmodel.h5'\n",
    "model_path = os.path.join(resultpath,modelname)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "best_model.save(model_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_reloaded = tf.keras.models.load_model(model_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The model has been reloaded. Let's reassure that this model has the same weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.all([np.all(x==y) for x,y in zip(best_model.get_weights(), model_reloaded.get_weights())])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Investigate model predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that the objects `models`, `best_model_fullytrained` and `best_model` that resulted from the mcfly functions are Keras objects. This means that you can use Keras functions on the objects, for example  `.predict`, (which when given the data, outputs the predictions for each sample) and `.evaluate` (which when given the data and the labels computes how well this model performs) . These functions are all documented in the [Keras documentation](https://www.tensorflow.org/api_docs/python/tf/keras). "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Inspect model predictions on validation data\n",
    "datasize = X_val.shape[0]\n",
    "probs = models[0][0].predict(X_val[:datasize,:,:],batch_size=1)  # here a different model was chosen!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's have a look at the [confusion matrix](https://en.wikipedia.org/wiki/Confusion_matrix)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "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>Badminton_Clear</th>\n",
       "      <th>Badminton_Smash</th>\n",
       "      <th>Squash_BackhandBoast</th>\n",
       "      <th>Squash_ForehandBoast</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Badminton_Clear</th>\n",
       "      <td>17</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Badminton_Smash</th>\n",
       "      <td>6</td>\n",
       "      <td>14</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Squash_BackhandBoast</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>17</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Squash_ForehandBoast</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      Badminton_Clear  Badminton_Smash  Squash_BackhandBoast  \\\n",
       "Badminton_Clear                    17                4                     0   \n",
       "Badminton_Smash                     6               14                     0   \n",
       "Squash_BackhandBoast                0                0                    17   \n",
       "Squash_ForehandBoast                0                0                     0   \n",
       "\n",
       "                      Squash_ForehandBoast  \n",
       "Badminton_Clear                          0  \n",
       "Badminton_Smash                          0  \n",
       "Squash_BackhandBoast                     0  \n",
       "Squash_ForehandBoast                    17  "
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#columns are predicted, rows are truth\n",
    "labels = lb.classes_\n",
    "\n",
    "predicted = probs.argmax(axis=1)\n",
    "y_index = y_val_binary.argmax(axis=1)\n",
    "confusion_matrix = pd.crosstab(pd.Series(y_index), pd.Series(predicted))\n",
    "confusion_matrix.index = [labels[i] for i in confusion_matrix.index]\n",
    "confusion_matrix.columns = [labels[i] for i in confusion_matrix.columns]\n",
    "confusion_matrix.reindex(columns=[l for l in labels], fill_value=0)\n",
    "confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}