ELU - Fashion MNIST.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8cf3be7d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.10.0'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "tf.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "4c1ec241",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.10.0'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "keras.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db8d1c8b",
   "metadata": {},
   "source": [
    "## Download the Data\n",
    "\n",
    "For this example we will work with the \"Fashin MNIST\" dataset, composed of thousands of $28 \\times 28$ images."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d53063b6",
   "metadata": {},
   "outputs": [],
   "source": [
    "fashion_mnist = keras.datasets.fashion_mnist\n",
    "(X_train_full, y_train_full), (X_test, y_test) = fashion_mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "2aac2bd1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000, 28, 28)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train_full.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "3baa0642",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 28, 28)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f86f2ea",
   "metadata": {},
   "source": [
    "Let's visualize the first image of our training set, so we have a better idea of how the data looks like. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4f0c3484",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib as mlp\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "sample_image = X_train_full[0]\n",
    "\n",
    "plt.imshow(sample_image, cmap=\"binary\")\n",
    "plt.axis(\"off\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45f80502",
   "metadata": {},
   "source": [
    "## Define Trainining and Validation Sets\n",
    "\n",
    "Now let's scale the data by dividing by the maximum value (255), so all the values are in the $[0, 1]$ range.\n",
    "\n",
    "We will create a validation set from our training set, composed of the first 5000 thousand images, the rest will be used for training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "537cb86a",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_valid, X_train = X_train_full[:5000] / 255.0, X_train_full[5000:] / 255.0\n",
    "y_valid, y_train = y_train_full[:5000], y_train_full[5000:]\n",
    "X_test = X_test / 255.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "c41020a1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(55000, 28, 28)\n",
      "(5000, 28, 28)\n"
     ]
    }
   ],
   "source": [
    "print(X_train.shape)\n",
    "print(X_valid.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dca45535",
   "metadata": {},
   "source": [
    "Let's define the names of our classes/categories."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e6ce9fcb",
   "metadata": {},
   "outputs": [],
   "source": [
    "class_names = [\"T-shirt/top\", \"Trouser\", \"Pullover\", \"Dress\", \n",
    "               \"Coat\", \"Sandal\", \"Shirt\", \"Sneaker\", \"Bag\", \"Ankle boot\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "569aafcd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Ankle boot'"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class_names[y_train_full[0]]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f4eb175b",
   "metadata": {},
   "source": [
    "## Create the  Model\n",
    "\n",
    "The input layer has 785 units, which represents the 28 x 28 pixels in the image. Let's define two hidden layers, with 300 and 100 neurons, respecitively. The output layer has 10 units, each representing a class; this output layer uses the `softmax`function to use a multinomial approach to handle multiple classes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "610a592e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " input (Flatten)             (None, 784)               0         \n",
      "                                                                 \n",
      " hidden1 (Dense)             (None, 300)               235500    \n",
      "                                                                 \n",
      " hidden2 (Dense)             (None, 100)               30100     \n",
      "                                                                 \n",
      " output (Dense)              (None, 10)                1010      \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 266,610\n",
      "Trainable params: 266,610\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Flatten(input_shape=[28, 28], name=\"input\"),\n",
    "    keras.layers.Dense(300, activation=\"elu\", kernel_initializer=\"he_normal\", name=\"hidden1\"),\n",
    "    keras.layers.Dense(100, activation=\"elu\", kernel_initializer=\"he_normal\", name=\"hidden2\"),\n",
    "    keras.layers.Dense(10, activation=\"softmax\", name=\"output\")\n",
    "])\n",
    "model.summary()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2666836d",
   "metadata": {},
   "source": [
    "We can get an array of the layers in our network with the `layers` propery."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "2ef9f71b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<keras.layers.reshaping.flatten.Flatten at 0x175ba1250>,\n",
       " <keras.layers.core.dense.Dense at 0x175ba1460>,\n",
       " <keras.layers.core.dense.Dense at 0x175ba14f0>,\n",
       " <keras.layers.core.dense.Dense at 0x175ba13d0>]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.layers"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0dcb5dd7",
   "metadata": {},
   "source": [
    "We can get an especific layer by index, and access the properties of that layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "4880756d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'hidden1'"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hidden1 = model.layers[1]\n",
    "hidden1.name"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e3baaf20",
   "metadata": {},
   "source": [
    "We also can get a layer by its name with the `get_layer` method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "4218d259",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.get_layer(\"hidden1\") is hidden1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "26693938",
   "metadata": {},
   "source": [
    "The `get_weights` method returns the weight parameters and bias terms of a layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "48f0caf8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[-0.04396542, -0.08827149,  0.05759838, ..., -0.05145448,\n",
       "         0.07065058,  0.0816386 ],\n",
       "       [ 0.03408444,  0.02899954,  0.03275677, ..., -0.00653306,\n",
       "         0.02851175,  0.07409915],\n",
       "       [-0.02346403, -0.06020518,  0.00028541, ..., -0.02028763,\n",
       "         0.02115271,  0.00734696],\n",
       "       ...,\n",
       "       [-0.07246459,  0.07841251, -0.02746085, ...,  0.00848408,\n",
       "         0.0367456 ,  0.01608934],\n",
       "       [ 0.06746966,  0.02942702,  0.0291285 , ..., -0.06629422,\n",
       "         0.02790489, -0.11107918],\n",
       "       [-0.09563631,  0.06010739,  0.01565216, ..., -0.00433768,\n",
       "         0.09096233, -0.06878874]], dtype=float32)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weights, biases = hidden1.get_weights()\n",
    "weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "87aef5ac",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(784, 300)\n",
      "(300,)\n"
     ]
    }
   ],
   "source": [
    "print(weights.shape)\n",
    "print(biases.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "79f61f7e",
   "metadata": {},
   "source": [
    "## Compile the Model\n",
    "\n",
    "In this step we specify the loss function, optimizers, and metrics for our network. For this example, we are using the categorical crossentroy loss function, since we are dealing with a multiclass categorical target variable. The `sgd` refers to the \"Stochastic Gradiend Descent\" algorithm, and we will use the accuracy for measuring the performance of our network. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "1ecb200e",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"sparse_categorical_crossentropy\",\n",
    "              optimizer=\"sgd\",\n",
    "              metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b877884d",
   "metadata": {},
   "source": [
    "## Adding Callbacks\n",
    "\n",
    "Callback are helpful to execute periodic specific tasks, for example, at the end of one epoch. In this case, we are adding an \"Early Stopping\" callback, indicating that if the validation score doesn't improve in 10 epochs, the training will stop and we will get the best solution computed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "a6ab0e72",
   "metadata": {},
   "outputs": [],
   "source": [
    "early_stopping_cb = keras.callbacks.EarlyStopping(patience=10, restore_best_weights=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6c739548",
   "metadata": {},
   "source": [
    "You can also define your own callbacks, here we are defining a callback that prints the ration between the validation score and the training score."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "d5ba10e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyCallback(keras.callbacks.Callback):\n",
    "    def on_epoch_end(self, epoch, logs):\n",
    "        print(f\"\\nval/train: {logs['val_loss'] / logs['loss']}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6cef0ff",
   "metadata": {},
   "source": [
    "## Training and Evaluating\n",
    "\n",
    "To train the network we need to specify the training and validation sets, as well as the number of epochs, callbacks ,and all the necessary parameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "9409a00b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "  50/1719 [..............................] - ETA: 1s - loss: 1.7011 - accuracy: 0.4350  "
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-02-06 15:12:08.320802: W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1719/1719 [==============================] - ETA: 0s - loss: 0.6372 - accuracy: 0.7823\n",
      "val/train: 0.8041150150061667\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.6372 - accuracy: 0.7823 - val_loss: 0.5124 - val_accuracy: 0.8140\n",
      "Epoch 2/100\n",
      "1672/1719 [============================>.] - ETA: 0s - loss: 0.4718 - accuracy: 0.8328\n",
      "val/train: 0.9212310463450071\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.4713 - accuracy: 0.8331 - val_loss: 0.4342 - val_accuracy: 0.8502\n",
      "Epoch 3/100\n",
      "1717/1719 [============================>.] - ETA: 0s - loss: 0.4374 - accuracy: 0.8445\n",
      "val/train: 0.9684584652973939\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.4373 - accuracy: 0.8445 - val_loss: 0.4235 - val_accuracy: 0.8550\n",
      "Epoch 4/100\n",
      "1713/1719 [============================>.] - ETA: 0s - loss: 0.4156 - accuracy: 0.8523\n",
      "val/train: 0.9843561449698514\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.4156 - accuracy: 0.8523 - val_loss: 0.4091 - val_accuracy: 0.8582\n",
      "Epoch 5/100\n",
      "1695/1719 [============================>.] - ETA: 0s - loss: 0.4010 - accuracy: 0.8571\n",
      "val/train: 0.9571375490605593\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.4005 - accuracy: 0.8575 - val_loss: 0.3833 - val_accuracy: 0.8652\n",
      "Epoch 6/100\n",
      "1711/1719 [============================>.] - ETA: 0s - loss: 0.3871 - accuracy: 0.8626\n",
      "val/train: 1.042639983584222\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3871 - accuracy: 0.8627 - val_loss: 0.4036 - val_accuracy: 0.8578\n",
      "Epoch 7/100\n",
      "1695/1719 [============================>.] - ETA: 0s - loss: 0.3762 - accuracy: 0.8658\n",
      "val/train: 1.0125933207872178\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3759 - accuracy: 0.8659 - val_loss: 0.3806 - val_accuracy: 0.8650\n",
      "Epoch 8/100\n",
      "1706/1719 [============================>.] - ETA: 0s - loss: 0.3690 - accuracy: 0.8675\n",
      "val/train: 1.0094866588478304\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3686 - accuracy: 0.8676 - val_loss: 0.3721 - val_accuracy: 0.8650\n",
      "Epoch 9/100\n",
      "1703/1719 [============================>.] - ETA: 0s - loss: 0.3603 - accuracy: 0.8700\n",
      "val/train: 1.0424044158718746\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3599 - accuracy: 0.8703 - val_loss: 0.3752 - val_accuracy: 0.8676\n",
      "Epoch 10/100\n",
      "1710/1719 [============================>.] - ETA: 0s - loss: 0.3526 - accuracy: 0.8731\n",
      "val/train: 0.998003572998025\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3529 - accuracy: 0.8730 - val_loss: 0.3522 - val_accuracy: 0.8710\n",
      "Epoch 11/100\n",
      "1692/1719 [============================>.] - ETA: 0s - loss: 0.3465 - accuracy: 0.8756\n",
      "val/train: 1.0600436743334674\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3457 - accuracy: 0.8758 - val_loss: 0.3665 - val_accuracy: 0.8648\n",
      "Epoch 12/100\n",
      "1714/1719 [============================>.] - ETA: 0s - loss: 0.3399 - accuracy: 0.8776\n",
      "val/train: 1.0470858428646526\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3396 - accuracy: 0.8777 - val_loss: 0.3556 - val_accuracy: 0.8746\n",
      "Epoch 13/100\n",
      "1688/1719 [============================>.] - ETA: 0s - loss: 0.3342 - accuracy: 0.8780\n",
      "val/train: 1.0333459007791965\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3338 - accuracy: 0.8782 - val_loss: 0.3449 - val_accuracy: 0.8720\n",
      "Epoch 14/100\n",
      "1693/1719 [============================>.] - ETA: 0s - loss: 0.3284 - accuracy: 0.8807\n",
      "val/train: 1.1131844658187868\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3288 - accuracy: 0.8806 - val_loss: 0.3660 - val_accuracy: 0.8690\n",
      "Epoch 15/100\n",
      "1688/1719 [============================>.] - ETA: 0s - loss: 0.3229 - accuracy: 0.8830\n",
      "val/train: 1.0358298905336016\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3228 - accuracy: 0.8830 - val_loss: 0.3344 - val_accuracy: 0.8812\n",
      "Epoch 16/100\n",
      "1693/1719 [============================>.] - ETA: 0s - loss: 0.3175 - accuracy: 0.8850\n",
      "val/train: 1.0491186954494596\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3178 - accuracy: 0.8851 - val_loss: 0.3334 - val_accuracy: 0.8756\n",
      "Epoch 17/100\n",
      "1706/1719 [============================>.] - ETA: 0s - loss: 0.3132 - accuracy: 0.8860\n",
      "val/train: 1.051582898663187\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3131 - accuracy: 0.8861 - val_loss: 0.3292 - val_accuracy: 0.8770\n",
      "Epoch 18/100\n",
      "1687/1719 [============================>.] - ETA: 0s - loss: 0.3084 - accuracy: 0.8882\n",
      "val/train: 1.1083049675232028\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3081 - accuracy: 0.8882 - val_loss: 0.3415 - val_accuracy: 0.8774\n",
      "Epoch 19/100\n",
      "1690/1719 [============================>.] - ETA: 0s - loss: 0.3045 - accuracy: 0.8889\n",
      "val/train: 1.0656720860828195\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.3046 - accuracy: 0.8888 - val_loss: 0.3246 - val_accuracy: 0.8792\n",
      "Epoch 20/100\n",
      "1674/1719 [============================>.] - ETA: 0s - loss: 0.2995 - accuracy: 0.8914\n",
      "val/train: 1.112477727924253\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2994 - accuracy: 0.8913 - val_loss: 0.3331 - val_accuracy: 0.8774\n",
      "Epoch 21/100\n",
      "1686/1719 [============================>.] - ETA: 0s - loss: 0.2965 - accuracy: 0.8926\n",
      "val/train: 1.2525181930418154\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2962 - accuracy: 0.8925 - val_loss: 0.3710 - val_accuracy: 0.8648\n",
      "Epoch 22/100\n",
      "1682/1719 [============================>.] - ETA: 0s - loss: 0.2930 - accuracy: 0.8927\n",
      "val/train: 1.1132837910722535\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2927 - accuracy: 0.8927 - val_loss: 0.3259 - val_accuracy: 0.8852\n",
      "Epoch 23/100\n",
      "1707/1719 [============================>.] - ETA: 0s - loss: 0.2888 - accuracy: 0.8953\n",
      "val/train: 1.1253381854759454\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2887 - accuracy: 0.8953 - val_loss: 0.3249 - val_accuracy: 0.8854\n",
      "Epoch 24/100\n",
      "1686/1719 [============================>.] - ETA: 0s - loss: 0.2846 - accuracy: 0.8968\n",
      "val/train: 1.134604847487407\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2851 - accuracy: 0.8965 - val_loss: 0.3234 - val_accuracy: 0.8804\n",
      "Epoch 25/100\n",
      "1686/1719 [============================>.] - ETA: 0s - loss: 0.2818 - accuracy: 0.8979\n",
      "val/train: 1.1333212237024997\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2820 - accuracy: 0.8977 - val_loss: 0.3196 - val_accuracy: 0.8858\n",
      "Epoch 26/100\n",
      "1706/1719 [============================>.] - ETA: 0s - loss: 0.2774 - accuracy: 0.8990\n",
      "val/train: 1.1834115623495027\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2776 - accuracy: 0.8990 - val_loss: 0.3285 - val_accuracy: 0.8834\n",
      "Epoch 27/100\n",
      "1704/1719 [============================>.] - ETA: 0s - loss: 0.2752 - accuracy: 0.8994\n",
      "val/train: 1.1565426309983775\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2754 - accuracy: 0.8993 - val_loss: 0.3185 - val_accuracy: 0.8840\n",
      "Epoch 28/100\n",
      "1681/1719 [============================>.] - ETA: 0s - loss: 0.2729 - accuracy: 0.9003\n",
      "val/train: 1.2374223903939536\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2719 - accuracy: 0.9006 - val_loss: 0.3364 - val_accuracy: 0.8752\n",
      "Epoch 29/100\n",
      "1708/1719 [============================>.] - ETA: 0s - loss: 0.2685 - accuracy: 0.9012\n",
      "val/train: 1.1923716858550744\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2687 - accuracy: 0.9013 - val_loss: 0.3203 - val_accuracy: 0.8866\n",
      "Epoch 30/100\n",
      "1704/1719 [============================>.] - ETA: 0s - loss: 0.2660 - accuracy: 0.9037\n",
      "val/train: 1.168777080164001\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2663 - accuracy: 0.9035 - val_loss: 0.3112 - val_accuracy: 0.8852\n",
      "Epoch 31/100\n",
      "1691/1719 [============================>.] - ETA: 0s - loss: 0.2639 - accuracy: 0.9040\n",
      "val/train: 1.21002455253341\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2639 - accuracy: 0.9040 - val_loss: 0.3193 - val_accuracy: 0.8856\n",
      "Epoch 32/100\n",
      "1685/1719 [============================>.] - ETA: 0s - loss: 0.2613 - accuracy: 0.9048\n",
      "val/train: 1.2106412987188848\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2606 - accuracy: 0.9052 - val_loss: 0.3155 - val_accuracy: 0.8808\n",
      "Epoch 33/100\n",
      "1681/1719 [============================>.] - ETA: 0s - loss: 0.2582 - accuracy: 0.9058\n",
      "val/train: 1.1859318986605678\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2586 - accuracy: 0.9056 - val_loss: 0.3067 - val_accuracy: 0.8886\n",
      "Epoch 34/100\n",
      "1691/1719 [============================>.] - ETA: 0s - loss: 0.2557 - accuracy: 0.9068\n",
      "val/train: 1.2006452084153425\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2558 - accuracy: 0.9067 - val_loss: 0.3071 - val_accuracy: 0.8902\n",
      "Epoch 35/100\n",
      "1705/1719 [============================>.] - ETA: 0s - loss: 0.2529 - accuracy: 0.9080\n",
      "val/train: 1.197954713622627\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2526 - accuracy: 0.9081 - val_loss: 0.3026 - val_accuracy: 0.8846\n",
      "Epoch 36/100\n",
      "1706/1719 [============================>.] - ETA: 0s - loss: 0.2496 - accuracy: 0.9100\n",
      "val/train: 1.2403700379190272\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2495 - accuracy: 0.9101 - val_loss: 0.3095 - val_accuracy: 0.8886\n",
      "Epoch 37/100\n",
      "1671/1719 [============================>.] - ETA: 0s - loss: 0.2472 - accuracy: 0.9112\n",
      "val/train: 1.2987276241960217\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2473 - accuracy: 0.9111 - val_loss: 0.3212 - val_accuracy: 0.8816\n",
      "Epoch 38/100\n",
      "1678/1719 [============================>.] - ETA: 0s - loss: 0.2447 - accuracy: 0.9113\n",
      "val/train: 1.2541571684859871\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2449 - accuracy: 0.9112 - val_loss: 0.3071 - val_accuracy: 0.8858\n",
      "Epoch 39/100\n",
      "1691/1719 [============================>.] - ETA: 0s - loss: 0.2435 - accuracy: 0.9119\n",
      "val/train: 1.2268457298142181\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2425 - accuracy: 0.9122 - val_loss: 0.2975 - val_accuracy: 0.8902\n",
      "Epoch 40/100\n",
      "1690/1719 [============================>.] - ETA: 0s - loss: 0.2399 - accuracy: 0.9141\n",
      "val/train: 1.2644984783316844\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2398 - accuracy: 0.9140 - val_loss: 0.3032 - val_accuracy: 0.8916\n",
      "Epoch 41/100\n",
      "1715/1719 [============================>.] - ETA: 0s - loss: 0.2387 - accuracy: 0.9130\n",
      "val/train: 1.2851671336358161\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2387 - accuracy: 0.9129 - val_loss: 0.3068 - val_accuracy: 0.8862\n",
      "Epoch 42/100\n",
      "1682/1719 [============================>.] - ETA: 0s - loss: 0.2358 - accuracy: 0.9147\n",
      "val/train: 1.3002147699303899\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2358 - accuracy: 0.9149 - val_loss: 0.3066 - val_accuracy: 0.8818\n",
      "Epoch 43/100\n",
      "1693/1719 [============================>.] - ETA: 0s - loss: 0.2338 - accuracy: 0.9158\n",
      "val/train: 1.3474690837120449\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2340 - accuracy: 0.9157 - val_loss: 0.3152 - val_accuracy: 0.8828\n",
      "Epoch 44/100\n",
      "1673/1719 [============================>.] - ETA: 0s - loss: 0.2315 - accuracy: 0.9163\n",
      "val/train: 1.296868502620452\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2313 - accuracy: 0.9165 - val_loss: 0.3000 - val_accuracy: 0.8894\n",
      "Epoch 45/100\n",
      "1707/1719 [============================>.] - ETA: 0s - loss: 0.2294 - accuracy: 0.9168\n",
      "val/train: 1.28467673316399\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2293 - accuracy: 0.9169 - val_loss: 0.2945 - val_accuracy: 0.8952\n",
      "Epoch 46/100\n",
      "1695/1719 [============================>.] - ETA: 0s - loss: 0.2281 - accuracy: 0.9169\n",
      "val/train: 1.2761939675038072\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2279 - accuracy: 0.9170 - val_loss: 0.2908 - val_accuracy: 0.8926\n",
      "Epoch 47/100\n",
      "1671/1719 [============================>.] - ETA: 0s - loss: 0.2255 - accuracy: 0.9181\n",
      "val/train: 1.470285989297077\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2250 - accuracy: 0.9183 - val_loss: 0.3308 - val_accuracy: 0.8792\n",
      "Epoch 48/100\n",
      "1695/1719 [============================>.] - ETA: 0s - loss: 0.2238 - accuracy: 0.9179\n",
      "val/train: 1.3375115905443802\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2237 - accuracy: 0.9181 - val_loss: 0.2992 - val_accuracy: 0.8902\n",
      "Epoch 49/100\n",
      "1698/1719 [============================>.] - ETA: 0s - loss: 0.2212 - accuracy: 0.9203\n",
      "val/train: 1.3479628216191546\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2212 - accuracy: 0.9202 - val_loss: 0.2982 - val_accuracy: 0.8878\n",
      "Epoch 50/100\n",
      "1697/1719 [============================>.] - ETA: 0s - loss: 0.2200 - accuracy: 0.9198\n",
      "val/train: 1.3451272565933148\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2198 - accuracy: 0.9199 - val_loss: 0.2957 - val_accuracy: 0.8926\n",
      "Epoch 51/100\n",
      "1672/1719 [============================>.] - ETA: 0s - loss: 0.2169 - accuracy: 0.9212\n",
      "val/train: 1.3922455257758823\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2168 - accuracy: 0.9211 - val_loss: 0.3019 - val_accuracy: 0.8892\n",
      "Epoch 52/100\n",
      "1691/1719 [============================>.] - ETA: 0s - loss: 0.2166 - accuracy: 0.9218\n",
      "val/train: 1.4008996120278272\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2160 - accuracy: 0.9220 - val_loss: 0.3026 - val_accuracy: 0.8896\n",
      "Epoch 53/100\n",
      "1682/1719 [============================>.] - ETA: 0s - loss: 0.2128 - accuracy: 0.9225\n",
      "val/train: 1.493810744577053\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2132 - accuracy: 0.9224 - val_loss: 0.3184 - val_accuracy: 0.8852\n",
      "Epoch 54/100\n",
      "1678/1719 [============================>.] - ETA: 0s - loss: 0.2118 - accuracy: 0.9226\n",
      "val/train: 1.4072991302478466\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2113 - accuracy: 0.9229 - val_loss: 0.2973 - val_accuracy: 0.8902\n",
      "Epoch 55/100\n",
      "1697/1719 [============================>.] - ETA: 0s - loss: 0.2089 - accuracy: 0.9241\n",
      "val/train: 1.4261369238683201\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2089 - accuracy: 0.9243 - val_loss: 0.2979 - val_accuracy: 0.8908\n",
      "Epoch 56/100\n",
      "1683/1719 [============================>.] - ETA: 0s - loss: 0.2070 - accuracy: 0.9257\n",
      "val/train: 1.4172050040320818\n",
      "1719/1719 [==============================] - 2s 1ms/step - loss: 0.2068 - accuracy: 0.9257 - val_loss: 0.2931 - val_accuracy: 0.8908\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(X_train, \n",
    "                    y_train, \n",
    "                    epochs=100, \n",
    "                    validation_data=(X_valid, y_valid),\n",
    "                    callbacks=[early_stopping_cb, MyCallback()]\n",
    "                   )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b537a308",
   "metadata": {},
   "source": [
    "With the `epoch` property we get a list with all the epochs that were completed. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "ed64ae54",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "56"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(history.epoch)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4472128b",
   "metadata": {},
   "source": [
    "The `steps` parameter refers to the number number of steps to complete an epoch, `batch_size * steps = n`. By default the batch size is 32."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "acb39dbf",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'verbose': 1, 'epochs': 100, 'steps': 1719}"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history.params"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9b7dd3ba",
   "metadata": {},
   "source": [
    "We can use the `history` property to obtain the score and loss of the training and validation sets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "c9672ea3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<function matplotlib.pyplot.show(close=None, block=None)>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 800x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "pd.DataFrame(history.history).plot(figsize=(8, 5))\n",
    "plt.grid(True)\n",
    "plt.gca().set_ylim(0, 1)\n",
    "plt.show"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6edb15f7",
   "metadata": {},
   "source": [
    "We use the `evaluate` method to measure the performance of a given data set. Below we display the loss and accuracy of the test set.\n",
    "\n",
    "Both, the accuracy and the loss are similar to the validation set, which is a sign that we aren't overfitting the training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "8f8ad911",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 [==============================] - 0s 655us/step - loss: 0.3269 - accuracy: 0.8816\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.3268720209598541, 0.881600022315979]"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(X_test, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "462e5156",
   "metadata": {},
   "source": [
    "The `predict` method returns the output of a forward step through the network with the given input. For this example, since we are using the `softmax` activation function; we get the probabilities to belong to each class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "40eb816e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 36ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.01, 0.  , 0.99],\n",
       "       [0.  , 0.  , 1.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  ],\n",
       "       [0.  , 1.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  ],\n",
       "       [0.  , 1.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  , 0.  ],\n",
       "       [0.12, 0.  , 0.03, 0.  , 0.01, 0.  , 0.84, 0.  , 0.  , 0.  ]],\n",
       "      dtype=float32)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_new = X_test[:5]\n",
    "y_proba = model.predict(X_new)\n",
    "y_proba.round(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "88537c84",
   "metadata": {},
   "source": [
    "Getting the class with the highest probability, we obtain the following."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "578a0297",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([9, 2, 1, 1, 6])"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "y_pred =np.argmax(y_proba, axis=1)\n",
    "y_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "82cc5ba1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['Ankle boot', 'Pullover', 'Trouser', 'Trouser', 'Shirt'],\n",
       "      dtype='<U11')"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(class_names)[y_pred]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2721defa",
   "metadata": {},
   "source": [
    "The results are identical to the actual values. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "3920eab4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([9, 2, 1, 1, 6], dtype=uint8)"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_new = y_test[:5]\n",
    "y_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f81bc80f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9b29c339",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": "3.8.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}