notebooks/06_04_Deep_Learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Chapitre 6 - Différentes utilisations du machine learning avec Python (4ème partie)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6.7 Le deep learning avec Keras\n",
    "\n",
    "### 6.7.2 Principe d’un réseau de neurone et première utilisation avec Keras\n",
    "\n",
    "Le package Keras que nous avons installé est basé sur une structure extrêmement\n",
    "simple à mettre en oeuvre. On commence par définir le type de réseau utilisé. Dans\n",
    "cet ouvrage, il s’agira de réseaux séquentiels."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Si nous prenons les données sur les télécommunications que nous avons utilisées\n",
    "précédemment, nous pouvons simplement utiliser Keras pour créer un réseau de neurones assez évolué (on suppose ici que les données ont été préparées comme\n",
    "dans la première partie de ce chapitre avec Scikit-Learn) :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Dropout\n",
    "from tensorflow.keras.layers import BatchNormalization, Activation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "x=pd.read_csv(\"../data/x_telecom.csv\",index_col=0)\n",
    "y=pd.read_csv(\"../data/y_telecom.csv\",index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "# on sépare nos données en apprentissage/validation avec Scikit-Learn\n",
    "x_train, x_test, y_train, y_test=train_test_split(x,y,test_size=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# on crée le modèle\n",
    "model = Sequential()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\ej\\miniconda3\\envs\\python-ds\\Lib\\site-packages\\keras\\src\\layers\\core\\dense.py:86: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.\n",
      "  super().__init__(activity_regularizer=activity_regularizer, **kwargs)\n"
     ]
    }
   ],
   "source": [
    "# on ajoute des couches avec des fonctions d’activation\n",
    "\n",
    "model.add(Dense(50, input_shape=(21,)))\n",
    "model.add(Activation(\"relu\"))\n",
    "# cette couche permet de normaliser les données\n",
    "model.add(BatchNormalization())\n",
    "# cette couche va faire en sorte d’éviter l’overfitting\n",
    "model.add(Dropout(0.1))\n",
    "# on ajoute un autre groupe de couches\n",
    "model.add(Dense(10))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Dropout(0.1))\n",
    "# on ajoute la couche de sortie\n",
    "model.add(Dense(1))\n",
    "model.add(Activation('sigmoid'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# on définit la fonction de perte et la fonction d’optiimsation\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 2ms/step - accuracy: 0.5067 - loss: 0.8147\n",
      "Epoch 2/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.6276 - loss: 0.6565 \n",
      "Epoch 3/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.7225 - loss: 0.5748 \n",
      "Epoch 4/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.7666 - loss: 0.5423 \n",
      "Epoch 5/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8113 - loss: 0.4808 \n",
      "Epoch 6/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8426 - loss: 0.4454 \n",
      "Epoch 7/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8707 - loss: 0.4045 \n",
      "Epoch 8/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8787 - loss: 0.3742 \n",
      "Epoch 9/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8794 - loss: 0.3636 \n",
      "Epoch 10/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.8842 - loss: 0.3361 \n",
      "Epoch 11/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9123 - loss: 0.2924 \n",
      "Epoch 12/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9146 - loss: 0.2785 \n",
      "Epoch 13/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9071 - loss: 0.2747\n",
      "Epoch 14/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9243 - loss: 0.2483 \n",
      "Epoch 15/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9291 - loss: 0.2389 \n",
      "Epoch 16/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9182 - loss: 0.2363 \n",
      "Epoch 17/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9163 - loss: 0.2281 \n",
      "Epoch 18/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9171 - loss: 0.2434\n",
      "Epoch 19/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9301 - loss: 0.2086 \n",
      "Epoch 20/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9372 - loss: 0.1987 \n",
      "Epoch 21/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9311 - loss: 0.2044 \n",
      "Epoch 22/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9306 - loss: 0.1930 \n",
      "Epoch 23/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9336 - loss: 0.2016 \n",
      "Epoch 24/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9420 - loss: 0.1907\n",
      "Epoch 25/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9340 - loss: 0.1880\n",
      "Epoch 26/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9384 - loss: 0.1929 \n",
      "Epoch 27/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9427 - loss: 0.1790 \n",
      "Epoch 28/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9364 - loss: 0.1746\n",
      "Epoch 29/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9325 - loss: 0.1950 \n",
      "Epoch 30/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9427 - loss: 0.1780 \n",
      "Epoch 31/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9302 - loss: 0.1982 \n",
      "Epoch 32/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9371 - loss: 0.1864\n",
      "Epoch 33/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9452 - loss: 0.1570 \n",
      "Epoch 34/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9329 - loss: 0.1967 \n",
      "Epoch 35/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9417 - loss: 0.1740 \n",
      "Epoch 36/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9384 - loss: 0.1816 \n",
      "Epoch 37/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9399 - loss: 0.1822 \n",
      "Epoch 38/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9513 - loss: 0.1573 \n",
      "Epoch 39/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9453 - loss: 0.1490 \n",
      "Epoch 40/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9461 - loss: 0.1608 \n",
      "Epoch 41/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9425 - loss: 0.1604 \n",
      "Epoch 42/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9505 - loss: 0.1518\n",
      "Epoch 43/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9494 - loss: 0.1455\n",
      "Epoch 44/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9516 - loss: 0.1487\n",
      "Epoch 45/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9365 - loss: 0.1726 \n",
      "Epoch 46/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9396 - loss: 0.1725 \n",
      "Epoch 47/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9496 - loss: 0.1516 \n",
      "Epoch 48/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9439 - loss: 0.1666 \n",
      "Epoch 49/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9505 - loss: 0.1497 \n",
      "Epoch 50/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9414 - loss: 0.1736 \n",
      "Epoch 51/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9523 - loss: 0.1612 \n",
      "Epoch 52/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9460 - loss: 0.1471 \n",
      "Epoch 53/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9458 - loss: 0.1556 \n",
      "Epoch 54/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9553 - loss: 0.1381 \n",
      "Epoch 55/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9451 - loss: 0.1626 \n",
      "Epoch 56/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9502 - loss: 0.1406 \n",
      "Epoch 57/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9540 - loss: 0.1497 \n",
      "Epoch 58/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9522 - loss: 0.1424 \n",
      "Epoch 59/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9545 - loss: 0.1382 \n",
      "Epoch 60/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9585 - loss: 0.1229 \n",
      "Epoch 61/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9485 - loss: 0.1444 \n",
      "Epoch 62/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9489 - loss: 0.1471 \n",
      "Epoch 63/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9587 - loss: 0.1237 \n",
      "Epoch 64/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9554 - loss: 0.1330 \n",
      "Epoch 65/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9517 - loss: 0.1412 \n",
      "Epoch 66/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9542 - loss: 0.1326\n",
      "Epoch 67/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9573 - loss: 0.1331 \n",
      "Epoch 68/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9565 - loss: 0.1312 \n",
      "Epoch 69/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9618 - loss: 0.1222 \n",
      "Epoch 70/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9653 - loss: 0.1185 \n",
      "Epoch 71/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9590 - loss: 0.1303 \n",
      "Epoch 72/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9588 - loss: 0.1187 \n",
      "Epoch 73/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9515 - loss: 0.1335 \n",
      "Epoch 74/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9533 - loss: 0.1560 \n",
      "Epoch 75/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9593 - loss: 0.1241 \n",
      "Epoch 76/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9606 - loss: 0.1352 \n",
      "Epoch 77/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9556 - loss: 0.1328 \n",
      "Epoch 78/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9617 - loss: 0.1136 \n",
      "Epoch 79/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9575 - loss: 0.1429 \n",
      "Epoch 80/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9611 - loss: 0.1249 \n",
      "Epoch 81/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9576 - loss: 0.1250 \n",
      "Epoch 82/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9595 - loss: 0.1297 \n",
      "Epoch 83/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9572 - loss: 0.1250 \n",
      "Epoch 84/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9665 - loss: 0.1065 \n",
      "Epoch 85/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9626 - loss: 0.1224 \n",
      "Epoch 86/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9550 - loss: 0.1347 \n",
      "Epoch 87/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9547 - loss: 0.1254 \n",
      "Epoch 88/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step - accuracy: 0.9522 - loss: 0.1389 \n",
      "Epoch 89/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9566 - loss: 0.1334 \n",
      "Epoch 90/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9608 - loss: 0.1173 \n",
      "Epoch 91/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9653 - loss: 0.1182 \n",
      "Epoch 92/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9664 - loss: 0.1118 \n",
      "Epoch 93/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9656 - loss: 0.1041 \n",
      "Epoch 94/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9675 - loss: 0.1032 \n",
      "Epoch 95/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9650 - loss: 0.1158 \n",
      "Epoch 96/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9579 - loss: 0.1353 \n",
      "Epoch 97/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9602 - loss: 0.1221 \n",
      "Epoch 98/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9630 - loss: 0.1088 \n",
      "Epoch 99/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9654 - loss: 0.1231 \n",
      "Epoch 100/100\n",
      "\u001b[1m27/27\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 1ms/step - accuracy: 0.9644 - loss: 0.1091 \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x269166097c0>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# on ajuste le modèle\n",
    "model.fit(x_train, y_train, batch_size=100, epochs=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m21/21\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[1.08628444e-01],\n",
       "       [1.26129771e-02],\n",
       "       [4.66779759e-03],\n",
       "       [7.27546662e-02],\n",
       "       [3.96000687e-03],\n",
       "       [3.16638537e-02],\n",
       "       [2.14160588e-02],\n",
       "       [2.75239404e-02],\n",
       "       [1.76696072e-03],\n",
       "       [1.96381006e-03],\n",
       "       [1.12803550e-02],\n",
       "       [2.05785222e-02],\n",
       "       [8.72267701e-04],\n",
       "       [9.23434738e-03],\n",
       "       [2.11154316e-02],\n",
       "       [9.19602998e-03],\n",
       "       [1.69141591e-02],\n",
       "       [4.21361960e-02],\n",
       "       [1.46912662e-02],\n",
       "       [6.62710052e-03],\n",
       "       [8.14505760e-03],\n",
       "       [2.15774179e-02],\n",
       "       [2.18369160e-02],\n",
       "       [1.89744681e-02],\n",
       "       [2.51232125e-02],\n",
       "       [2.05599871e-02],\n",
       "       [2.33918186e-02],\n",
       "       [9.68107209e-03],\n",
       "       [7.25798635e-03],\n",
       "       [3.58934863e-03],\n",
       "       [9.67160892e-03],\n",
       "       [6.33288920e-02],\n",
       "       [4.63237101e-03],\n",
       "       [2.72004213e-02],\n",
       "       [9.64996591e-03],\n",
       "       [1.57859456e-02],\n",
       "       [2.79434249e-02],\n",
       "       [9.84488148e-03],\n",
       "       [2.17328444e-01],\n",
       "       [6.99033029e-03],\n",
       "       [6.12041680e-03],\n",
       "       [2.67993682e-03],\n",
       "       [9.06997621e-01],\n",
       "       [9.55774169e-03],\n",
       "       [4.52880980e-03],\n",
       "       [1.09051242e-02],\n",
       "       [4.32621734e-03],\n",
       "       [2.13277843e-02],\n",
       "       [1.69097446e-03],\n",
       "       [6.19732440e-01],\n",
       "       [1.63073651e-02],\n",
       "       [8.55819602e-03],\n",
       "       [3.41057923e-04],\n",
       "       [4.38200869e-02],\n",
       "       [9.74684060e-02],\n",
       "       [5.50328661e-03],\n",
       "       [3.59104052e-02],\n",
       "       [3.30286287e-02],\n",
       "       [1.36825573e-02],\n",
       "       [9.66929365e-03],\n",
       "       [9.98395681e-01],\n",
       "       [3.05905752e-02],\n",
       "       [1.55522749e-02],\n",
       "       [8.51574820e-04],\n",
       "       [7.60820881e-03],\n",
       "       [9.44065973e-02],\n",
       "       [1.87762659e-02],\n",
       "       [3.59187685e-02],\n",
       "       [4.54183249e-03],\n",
       "       [2.69915368e-02],\n",
       "       [3.60977352e-02],\n",
       "       [6.63544284e-03],\n",
       "       [1.10144541e-02],\n",
       "       [6.17001802e-02],\n",
       "       [6.14367705e-03],\n",
       "       [2.36428827e-02],\n",
       "       [7.34683825e-03],\n",
       "       [9.23114549e-03],\n",
       "       [3.73758152e-02],\n",
       "       [9.10526514e-03],\n",
       "       [5.67379221e-02],\n",
       "       [4.74725571e-03],\n",
       "       [3.21039883e-03],\n",
       "       [2.71744300e-02],\n",
       "       [9.80788656e-03],\n",
       "       [8.61136056e-03],\n",
       "       [1.18136778e-02],\n",
       "       [6.49488112e-03],\n",
       "       [2.59283110e-02],\n",
       "       [4.10662824e-03],\n",
       "       [2.95921247e-02],\n",
       "       [9.95089829e-01],\n",
       "       [9.66155291e-01],\n",
       "       [1.68344076e-03],\n",
       "       [9.99999940e-01],\n",
       "       [1.97017118e-02],\n",
       "       [3.22034792e-03],\n",
       "       [2.85193976e-02],\n",
       "       [1.38173287e-03],\n",
       "       [7.88221061e-01],\n",
       "       [7.98721856e-04],\n",
       "       [9.16697085e-03],\n",
       "       [5.41937677e-03],\n",
       "       [3.62677383e-03],\n",
       "       [1.96171790e-01],\n",
       "       [2.23922078e-02],\n",
       "       [7.05842581e-03],\n",
       "       [2.06267629e-02],\n",
       "       [2.66385853e-01],\n",
       "       [4.37015295e-01],\n",
       "       [2.27433555e-02],\n",
       "       [1.28347632e-02],\n",
       "       [4.48331004e-03],\n",
       "       [3.91645450e-03],\n",
       "       [2.14669555e-02],\n",
       "       [1.82866270e-03],\n",
       "       [6.39360677e-03],\n",
       "       [8.51803944e-02],\n",
       "       [3.56519106e-03],\n",
       "       [9.99973357e-01],\n",
       "       [4.87285061e-03],\n",
       "       [1.01044569e-02],\n",
       "       [1.38255442e-03],\n",
       "       [2.21425411e-03],\n",
       "       [2.16926215e-03],\n",
       "       [7.17744604e-03],\n",
       "       [6.29183021e-04],\n",
       "       [1.66989006e-02],\n",
       "       [3.30547616e-03],\n",
       "       [3.37688148e-01],\n",
       "       [1.98034458e-02],\n",
       "       [1.70188528e-02],\n",
       "       [4.55116760e-03],\n",
       "       [1.03332670e-02],\n",
       "       [6.72290742e-01],\n",
       "       [5.80466411e-04],\n",
       "       [5.36131322e-01],\n",
       "       [3.03551718e-03],\n",
       "       [1.41983582e-02],\n",
       "       [7.34309759e-03],\n",
       "       [1.58646610e-02],\n",
       "       [2.47616824e-02],\n",
       "       [9.99999762e-01],\n",
       "       [1.56034539e-02],\n",
       "       [6.41727215e-03],\n",
       "       [9.99785542e-01],\n",
       "       [2.27993503e-02],\n",
       "       [5.95426653e-03],\n",
       "       [6.77594319e-02],\n",
       "       [1.41415058e-03],\n",
       "       [3.01962737e-02],\n",
       "       [2.85650580e-03],\n",
       "       [5.51073015e-01],\n",
       "       [1.13698132e-01],\n",
       "       [3.39189172e-02],\n",
       "       [2.15219590e-03],\n",
       "       [4.67571393e-02],\n",
       "       [3.88783426e-03],\n",
       "       [5.84265799e-04],\n",
       "       [2.23158509e-03],\n",
       "       [3.05990782e-03],\n",
       "       [2.31052935e-02],\n",
       "       [4.38639373e-02],\n",
       "       [2.33340897e-02],\n",
       "       [4.86847758e-03],\n",
       "       [1.25089753e-02],\n",
       "       [1.14865815e-02],\n",
       "       [3.04978574e-03],\n",
       "       [7.17576733e-03],\n",
       "       [4.91102934e-02],\n",
       "       [9.97560263e-01],\n",
       "       [2.38616159e-03],\n",
       "       [3.06853373e-02],\n",
       "       [1.51069500e-02],\n",
       "       [4.89011593e-03],\n",
       "       [3.37055065e-02],\n",
       "       [2.12060544e-03],\n",
       "       [8.91083945e-03],\n",
       "       [9.99398530e-01],\n",
       "       [5.04102418e-03],\n",
       "       [1.05735371e-02],\n",
       "       [2.56591593e-03],\n",
       "       [3.47330188e-03],\n",
       "       [7.81385880e-03],\n",
       "       [7.54886307e-03],\n",
       "       [4.37029218e-03],\n",
       "       [2.80118943e-03],\n",
       "       [2.50561512e-03],\n",
       "       [9.54290212e-04],\n",
       "       [4.45431247e-02],\n",
       "       [1.67611218e-03],\n",
       "       [1.59893304e-01],\n",
       "       [2.25020815e-02],\n",
       "       [3.74789815e-03],\n",
       "       [3.39891785e-03],\n",
       "       [6.91720918e-02],\n",
       "       [3.62249953e-03],\n",
       "       [1.06704719e-02],\n",
       "       [8.18316732e-03],\n",
       "       [1.46841519e-02],\n",
       "       [1.20744165e-02],\n",
       "       [4.48648930e-02],\n",
       "       [4.24745539e-03],\n",
       "       [2.96232253e-02],\n",
       "       [2.69963173e-03],\n",
       "       [5.37607539e-03],\n",
       "       [7.79297343e-03],\n",
       "       [4.56007943e-02],\n",
       "       [1.00000000e+00],\n",
       "       [8.75362158e-01],\n",
       "       [2.28849798e-02],\n",
       "       [1.38660595e-01],\n",
       "       [1.71182794e-03],\n",
       "       [6.29316866e-01],\n",
       "       [4.14876267e-03],\n",
       "       [1.05491458e-02],\n",
       "       [2.73905112e-03],\n",
       "       [2.48079794e-03],\n",
       "       [1.16163343e-01],\n",
       "       [1.34463767e-02],\n",
       "       [3.14515866e-02],\n",
       "       [1.13737732e-02],\n",
       "       [2.18179747e-02],\n",
       "       [7.44944334e-01],\n",
       "       [1.77338300e-03],\n",
       "       [1.07912510e-03],\n",
       "       [7.53360451e-04],\n",
       "       [4.40789610e-01],\n",
       "       [6.63474854e-03],\n",
       "       [1.00000000e+00],\n",
       "       [1.31218946e-02],\n",
       "       [4.49417010e-02],\n",
       "       [9.38560069e-03],\n",
       "       [2.62251180e-02],\n",
       "       [1.00495983e-02],\n",
       "       [6.88186428e-03],\n",
       "       [5.81088476e-03],\n",
       "       [1.85856193e-01],\n",
       "       [8.87762010e-03],\n",
       "       [1.94693413e-02],\n",
       "       [5.64820402e-05],\n",
       "       [9.43536937e-01],\n",
       "       [8.86768222e-01],\n",
       "       [2.23510399e-01],\n",
       "       [6.71929261e-03],\n",
       "       [9.98112559e-01],\n",
       "       [9.91807163e-01],\n",
       "       [1.50225842e-02],\n",
       "       [1.09072234e-02],\n",
       "       [1.49713876e-03],\n",
       "       [2.15984713e-02],\n",
       "       [3.08932201e-03],\n",
       "       [4.25987281e-02],\n",
       "       [1.44047262e-02],\n",
       "       [5.78597141e-03],\n",
       "       [9.46053118e-03],\n",
       "       [3.88998818e-03],\n",
       "       [1.52863730e-02],\n",
       "       [1.03080878e-02],\n",
       "       [7.70715903e-03],\n",
       "       [4.63763550e-02],\n",
       "       [6.98347762e-02],\n",
       "       [9.67480708e-03],\n",
       "       [7.13025127e-03],\n",
       "       [1.27919512e-02],\n",
       "       [1.34155139e-01],\n",
       "       [1.13041170e-01],\n",
       "       [1.44986315e-02],\n",
       "       [9.48679354e-03],\n",
       "       [4.63166647e-02],\n",
       "       [7.33007537e-03],\n",
       "       [2.56973684e-01],\n",
       "       [7.68757582e-01],\n",
       "       [2.91725009e-04],\n",
       "       [4.54681693e-03],\n",
       "       [7.98843861e-01],\n",
       "       [7.09079904e-03],\n",
       "       [5.16337305e-02],\n",
       "       [2.41083093e-03],\n",
       "       [2.18146443e-02],\n",
       "       [9.20204166e-03],\n",
       "       [3.98566648e-02],\n",
       "       [6.42940879e-01],\n",
       "       [4.19235032e-04],\n",
       "       [1.36810839e-02],\n",
       "       [4.06922353e-03],\n",
       "       [1.55515028e-02],\n",
       "       [1.34943007e-02],\n",
       "       [3.89850792e-03],\n",
       "       [8.00304639e-04],\n",
       "       [2.97769741e-03],\n",
       "       [8.97605047e-02],\n",
       "       [1.05392724e-01],\n",
       "       [3.31409983e-02],\n",
       "       [2.38183839e-03],\n",
       "       [2.28458596e-03],\n",
       "       [5.88843366e-03],\n",
       "       [3.77433263e-02],\n",
       "       [3.19347680e-02],\n",
       "       [4.60838638e-02],\n",
       "       [8.72484326e-01],\n",
       "       [3.74781829e-03],\n",
       "       [2.01059263e-02],\n",
       "       [8.17791939e-01],\n",
       "       [5.84854977e-03],\n",
       "       [1.00699134e-01],\n",
       "       [8.44688434e-03],\n",
       "       [9.87964682e-03],\n",
       "       [2.95922812e-03],\n",
       "       [2.74638622e-03],\n",
       "       [1.75187865e-03],\n",
       "       [1.44047262e-02],\n",
       "       [1.36824017e-02],\n",
       "       [9.69972927e-03],\n",
       "       [9.13865685e-01],\n",
       "       [1.00000000e+00],\n",
       "       [9.76779871e-03],\n",
       "       [3.30832005e-01],\n",
       "       [2.45162728e-03],\n",
       "       [4.30049486e-02],\n",
       "       [2.97211140e-04],\n",
       "       [1.56999729e-03],\n",
       "       [1.02157937e-02],\n",
       "       [9.99989152e-01],\n",
       "       [2.58186404e-02],\n",
       "       [1.22300945e-02],\n",
       "       [1.30640611e-03],\n",
       "       [1.70878954e-02],\n",
       "       [8.81700404e-03],\n",
       "       [1.31429022e-03],\n",
       "       [8.21956873e-01],\n",
       "       [1.58691604e-04],\n",
       "       [1.40106842e-01],\n",
       "       [6.85524382e-03],\n",
       "       [2.90712081e-02],\n",
       "       [3.61562483e-02],\n",
       "       [4.45452891e-03],\n",
       "       [3.81182972e-03],\n",
       "       [4.72968876e-01],\n",
       "       [5.01948548e-03],\n",
       "       [1.28274607e-02],\n",
       "       [1.33009453e-03],\n",
       "       [1.27299810e-02],\n",
       "       [1.68027080e-04],\n",
       "       [6.91056624e-03],\n",
       "       [2.80098952e-02],\n",
       "       [2.63421331e-02],\n",
       "       [2.08079461e-02],\n",
       "       [6.11495133e-03],\n",
       "       [8.27603042e-03],\n",
       "       [3.95487342e-03],\n",
       "       [1.00947563e-02],\n",
       "       [1.26858708e-02],\n",
       "       [6.20708382e-03],\n",
       "       [6.92222267e-04],\n",
       "       [3.04183667e-03],\n",
       "       [2.64114072e-03],\n",
       "       [3.69679369e-03],\n",
       "       [1.13061303e-02],\n",
       "       [1.07670330e-01],\n",
       "       [4.87631327e-03],\n",
       "       [2.97376700e-03],\n",
       "       [2.99371686e-03],\n",
       "       [5.67913707e-03],\n",
       "       [1.00000000e+00],\n",
       "       [4.57738265e-02],\n",
       "       [1.60650741e-02],\n",
       "       [1.97306741e-02],\n",
       "       [1.25131281e-02],\n",
       "       [1.85847413e-02],\n",
       "       [1.72240306e-02],\n",
       "       [2.64412127e-02],\n",
       "       [9.99179780e-01],\n",
       "       [2.03094900e-01],\n",
       "       [8.81224812e-04],\n",
       "       [3.04755243e-03],\n",
       "       [9.83500659e-01],\n",
       "       [4.55526374e-02],\n",
       "       [4.38026696e-01],\n",
       "       [5.26961274e-02],\n",
       "       [1.49055654e-02],\n",
       "       [2.63381526e-02],\n",
       "       [9.99999523e-01],\n",
       "       [7.77232368e-03],\n",
       "       [8.66978168e-02],\n",
       "       [6.21792348e-03],\n",
       "       [2.16992587e-01],\n",
       "       [3.32423579e-03],\n",
       "       [1.84950395e-03],\n",
       "       [6.23655214e-04],\n",
       "       [3.47678293e-03],\n",
       "       [4.86656092e-03],\n",
       "       [2.88171018e-03],\n",
       "       [8.83218925e-03],\n",
       "       [1.66740734e-03],\n",
       "       [1.44047262e-02],\n",
       "       [8.61761998e-03],\n",
       "       [9.27190065e-01],\n",
       "       [1.24880811e-03],\n",
       "       [2.32485570e-02],\n",
       "       [2.74868663e-02],\n",
       "       [7.75969541e-03],\n",
       "       [4.32324141e-01],\n",
       "       [8.00719857e-01],\n",
       "       [9.97815013e-01],\n",
       "       [8.33906978e-03],\n",
       "       [9.99996543e-01],\n",
       "       [1.23607051e-02],\n",
       "       [2.06567291e-02],\n",
       "       [6.56176184e-04],\n",
       "       [1.96422287e-03],\n",
       "       [2.67905509e-03],\n",
       "       [2.34176517e-02],\n",
       "       [1.02120362e-01],\n",
       "       [1.41938534e-02],\n",
       "       [3.74413550e-01],\n",
       "       [1.69714130e-02],\n",
       "       [3.02361064e-02],\n",
       "       [4.20329906e-03],\n",
       "       [9.99931276e-01],\n",
       "       [1.21804416e-01],\n",
       "       [2.17676745e-03],\n",
       "       [4.60776687e-03],\n",
       "       [4.35242942e-03],\n",
       "       [9.95655954e-01],\n",
       "       [2.27895789e-02],\n",
       "       [4.04483732e-03],\n",
       "       [7.30884029e-03],\n",
       "       [2.46267579e-02],\n",
       "       [1.54108539e-01],\n",
       "       [8.65084052e-01],\n",
       "       [2.93885153e-02],\n",
       "       [8.65374785e-03],\n",
       "       [3.26454313e-03],\n",
       "       [2.75852177e-02],\n",
       "       [7.01307058e-01],\n",
       "       [3.72137837e-02],\n",
       "       [1.75886843e-02],\n",
       "       [4.29206947e-03],\n",
       "       [2.32640021e-02],\n",
       "       [7.53499031e-01],\n",
       "       [1.27243837e-02],\n",
       "       [1.32742301e-02],\n",
       "       [6.08526289e-01],\n",
       "       [1.62279152e-03],\n",
       "       [9.99999106e-01],\n",
       "       [5.77228609e-03],\n",
       "       [2.23116600e-03],\n",
       "       [4.95114654e-01],\n",
       "       [5.54059930e-02],\n",
       "       [8.05480685e-03],\n",
       "       [9.81745310e-03],\n",
       "       [9.86689687e-01],\n",
       "       [1.92500334e-02],\n",
       "       [1.05756306e-04],\n",
       "       [1.17522851e-02],\n",
       "       [2.43710959e-03],\n",
       "       [1.01912571e-02],\n",
       "       [1.01094879e-03],\n",
       "       [3.66976373e-02],\n",
       "       [6.50166580e-03],\n",
       "       [5.09126522e-02],\n",
       "       [4.14278172e-02],\n",
       "       [3.46901640e-02],\n",
       "       [8.27861428e-01],\n",
       "       [1.53170223e-03],\n",
       "       [1.07436739e-02],\n",
       "       [1.88065565e-03],\n",
       "       [9.96463653e-03],\n",
       "       [1.84614630e-03],\n",
       "       [1.70984790e-02],\n",
       "       [7.71653326e-03],\n",
       "       [3.56861949e-03],\n",
       "       [5.54171717e-03],\n",
       "       [1.08659612e-02],\n",
       "       [1.24330614e-02],\n",
       "       [5.00671007e-03],\n",
       "       [1.45863211e-02],\n",
       "       [1.24013741e-02],\n",
       "       [2.81816516e-02],\n",
       "       [3.71393263e-02],\n",
       "       [2.12646695e-03],\n",
       "       [1.18644512e-03],\n",
       "       [9.19864774e-01],\n",
       "       [1.13613248e-01],\n",
       "       [1.15592942e-01],\n",
       "       [1.04792044e-02],\n",
       "       [1.97877409e-03],\n",
       "       [7.02942461e-02],\n",
       "       [4.97436430e-03],\n",
       "       [2.37364825e-02],\n",
       "       [1.88815070e-03],\n",
       "       [1.23201748e-02],\n",
       "       [2.02562530e-02],\n",
       "       [4.13276672e-01],\n",
       "       [6.56709168e-03],\n",
       "       [9.99959588e-01],\n",
       "       [1.78141345e-03],\n",
       "       [6.09202869e-02],\n",
       "       [1.03018964e-02],\n",
       "       [2.87082978e-03],\n",
       "       [2.33474635e-02],\n",
       "       [1.45254256e-02],\n",
       "       [8.09720993e-01],\n",
       "       [9.93461311e-01],\n",
       "       [9.58653316e-02],\n",
       "       [1.75037328e-02],\n",
       "       [1.66039653e-02],\n",
       "       [1.05128698e-02],\n",
       "       [1.74490381e-02],\n",
       "       [1.00000000e+00],\n",
       "       [1.89358159e-03],\n",
       "       [1.15647223e-02],\n",
       "       [9.04773399e-02],\n",
       "       [3.27513665e-02],\n",
       "       [3.74496158e-04],\n",
       "       [6.90757576e-03],\n",
       "       [9.99999642e-01],\n",
       "       [9.78245283e-04],\n",
       "       [3.33509478e-03],\n",
       "       [3.40394164e-03],\n",
       "       [2.34293751e-02],\n",
       "       [7.36074802e-03],\n",
       "       [1.09349526e-01],\n",
       "       [3.23091284e-03],\n",
       "       [3.05238785e-03],\n",
       "       [2.21725293e-02],\n",
       "       [2.06519896e-03],\n",
       "       [4.59951311e-02],\n",
       "       [9.98147011e-01],\n",
       "       [5.62797748e-02],\n",
       "       [4.21845209e-04],\n",
       "       [9.96248901e-01],\n",
       "       [8.86943005e-03],\n",
       "       [1.51114794e-03],\n",
       "       [1.18685784e-02],\n",
       "       [1.00227147e-01],\n",
       "       [3.37077975e-02],\n",
       "       [8.16802587e-03],\n",
       "       [1.87193863e-02],\n",
       "       [2.20050497e-04],\n",
       "       [4.99552768e-03],\n",
       "       [1.37667276e-03],\n",
       "       [5.73120080e-03],\n",
       "       [3.02479744e-01],\n",
       "       [5.84189966e-03],\n",
       "       [2.46536545e-03],\n",
       "       [8.53452599e-04],\n",
       "       [4.04934119e-03],\n",
       "       [3.21119800e-02],\n",
       "       [1.58455018e-02],\n",
       "       [9.99996006e-01],\n",
       "       [1.50251118e-02],\n",
       "       [5.13013825e-03],\n",
       "       [3.23337852e-03],\n",
       "       [6.20867983e-02],\n",
       "       [7.67010171e-03],\n",
       "       [1.22624235e-02],\n",
       "       [2.21580509e-02],\n",
       "       [2.67328718e-03],\n",
       "       [2.73955576e-02],\n",
       "       [4.42759395e-02],\n",
       "       [2.46148580e-03],\n",
       "       [3.87524720e-04],\n",
       "       [6.84029702e-03],\n",
       "       [3.84486616e-02],\n",
       "       [3.50137800e-03],\n",
       "       [1.00000000e+00],\n",
       "       [5.03633590e-03],\n",
       "       [3.58135533e-03],\n",
       "       [1.64009770e-03],\n",
       "       [6.21891581e-02],\n",
       "       [1.36726825e-02],\n",
       "       [1.06715085e-02],\n",
       "       [2.04496887e-02],\n",
       "       [1.02330064e-02],\n",
       "       [6.70142646e-04],\n",
       "       [2.10510209e-01],\n",
       "       [1.11686643e-02],\n",
       "       [1.10150520e-02],\n",
       "       [5.92098478e-03],\n",
       "       [6.78025605e-03],\n",
       "       [2.47859908e-03],\n",
       "       [6.90184720e-03],\n",
       "       [1.32345352e-02],\n",
       "       [7.96707813e-03],\n",
       "       [9.34587978e-03],\n",
       "       [2.03505773e-02],\n",
       "       [2.79847626e-03],\n",
       "       [7.61774834e-03],\n",
       "       [4.97491568e-01],\n",
       "       [5.60272671e-02],\n",
       "       [4.22528898e-03],\n",
       "       [3.37014033e-04],\n",
       "       [3.38627282e-03],\n",
       "       [7.59913549e-02],\n",
       "       [2.04139831e-03],\n",
       "       [8.20062637e-01],\n",
       "       [1.17324749e-02],\n",
       "       [2.08914583e-03],\n",
       "       [5.92856482e-03],\n",
       "       [6.15531951e-03],\n",
       "       [9.80044622e-03],\n",
       "       [1.56504498e-03],\n",
       "       [8.67474638e-03],\n",
       "       [6.83342963e-02],\n",
       "       [2.10262705e-02],\n",
       "       [1.13550399e-03],\n",
       "       [3.83392745e-03],\n",
       "       [4.49345224e-02],\n",
       "       [1.03705406e-01],\n",
       "       [2.22234186e-02],\n",
       "       [1.82126544e-03],\n",
       "       [9.97948408e-01],\n",
       "       [4.39587020e-04],\n",
       "       [3.64361554e-02],\n",
       "       [1.07390096e-03],\n",
       "       [7.17740282e-02],\n",
       "       [5.36836823e-03],\n",
       "       [5.77826379e-03],\n",
       "       [1.81185908e-03],\n",
       "       [4.97821765e-03],\n",
       "       [3.14389579e-02],\n",
       "       [5.83553361e-03],\n",
       "       [9.94721428e-03],\n",
       "       [1.05655249e-02],\n",
       "       [1.69509482e-02],\n",
       "       [9.90041196e-01],\n",
       "       [4.28657746e-03],\n",
       "       [5.10006137e-02],\n",
       "       [8.60148668e-03],\n",
       "       [3.37575749e-02],\n",
       "       [1.07530234e-02],\n",
       "       [9.11509469e-02],\n",
       "       [1.63063463e-02],\n",
       "       [8.75698216e-03],\n",
       "       [1.53616886e-04],\n",
       "       [1.52684958e-03],\n",
       "       [1.45353954e-02],\n",
       "       [3.20331729e-03],\n",
       "       [1.44047262e-02],\n",
       "       [1.06374016e-02],\n",
       "       [9.99994576e-01],\n",
       "       [6.00624159e-02],\n",
       "       [9.06621863e-04],\n",
       "       [5.73070571e-02],\n",
       "       [3.76868173e-02],\n",
       "       [2.68448051e-02],\n",
       "       [3.07925016e-01],\n",
       "       [5.92153810e-04],\n",
       "       [1.65456012e-02],\n",
       "       [1.07920729e-03],\n",
       "       [9.16427188e-03],\n",
       "       [8.92417412e-03],\n",
       "       [8.58496502e-03],\n",
       "       [4.02651582e-04],\n",
       "       [1.14712091e-02],\n",
       "       [9.99890447e-01],\n",
       "       [2.69923925e-01],\n",
       "       [8.63892771e-03],\n",
       "       [9.91200626e-01],\n",
       "       [3.29491645e-02],\n",
       "       [8.30391515e-03],\n",
       "       [9.51212585e-01],\n",
       "       [1.26817627e-02],\n",
       "       [2.48922724e-02],\n",
       "       [4.36813058e-03]], dtype=float32)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m21/21\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 2ms/step \n",
      "AUC pour RN : 0.9234279321714555\n"
     ]
    }
   ],
   "source": [
    "# on calcule l’AUC pour les données de validation\n",
    "from sklearn.metrics import roc_auc_score\n",
    "print(\"AUC pour RN :\", roc_auc_score(y_test,\n",
    "                                     model.predict(x_test).reshape(-1)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Les réseaux de neurones sont des méthodes extrêmement efficaces à condition\n",
    "de bien les paramétrer ce qui peut s‘avérer très difficile."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.7.3 Le deep learning et les réseaux de neurones à convolutions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On va maintenant travailler sur des images. On prendra les données fashion-mnist :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten\n",
    "from tensorflow.keras.layers import Conv2D, MaxPool2D\n",
    "from tensorflow.keras.datasets import fashion_mnist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# récupération des données\n",
    "(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()\n",
    "\n",
    "# les données sont transformées pour passer sous un format d’image\n",
    "# standard avec des valeurs entre 0 et 1\n",
    "x_train = (x_train/255).reshape(x_train.shape[0], 28, 28,1)\n",
    "x_test = (x_test/255).reshape(x_test.shape[0], 28, 28,1)\n",
    "\n",
    "# passage à 10 colonnes pour y(une par modalité)\n",
    "y_train = to_categorical(y_train, 10)\n",
    "y_test = to_categorical(y_test, 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\ej\\miniconda3\\envs\\python-ds\\Lib\\site-packages\\keras\\src\\layers\\convolutional\\base_conv.py:99: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.\n",
      "  super().__init__(\n"
     ]
    }
   ],
   "source": [
    "# construction du modèle\n",
    "model = Sequential()\n",
    "model.add(Conv2D(32, (3, 3), activation = 'relu', input_shape = (28,28,1)))\n",
    "model.add(Conv2D(32, (3, 3), activation = 'relu'))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Dropout(0.25))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(128, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(10, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# compilation du modèle\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 10ms/step - accuracy: 0.7407 - loss: 0.7360\n",
      "Epoch 2/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 10ms/step - accuracy: 0.8714 - loss: 0.3572\n",
      "Epoch 3/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 10ms/step - accuracy: 0.8898 - loss: 0.3104\n",
      "Epoch 4/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 15ms/step - accuracy: 0.8976 - loss: 0.2785\n",
      "Epoch 5/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 15ms/step - accuracy: 0.9058 - loss: 0.2556\n",
      "Epoch 6/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 16ms/step - accuracy: 0.9129 - loss: 0.2385\n",
      "Epoch 7/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 16ms/step - accuracy: 0.9194 - loss: 0.2173\n",
      "Epoch 8/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 16ms/step - accuracy: 0.9242 - loss: 0.2065\n",
      "Epoch 9/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 19ms/step - accuracy: 0.9276 - loss: 0.1941\n",
      "Epoch 10/10\n",
      "\u001b[1m600/600\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 17ms/step - accuracy: 0.9315 - loss: 0.1833\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x26917b046b0>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# ajustement du modèle\n",
    "model.fit(x_train, y_train, batch_size=100, epochs=10, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# evaluation du modèle\n",
    "score = model.evaluate(x_test, y_test, verbose=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous avons appliqué deux couches de convolutions à 32 neurones avec une taille\n",
    "de fenêtre de 3 par 3. Puis une couche de pooling avec une réduction sur des fenêtres\n",
    "de 2 par 2. Nous obtenons les résultats suivants :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pourcentage de bien classées : 0.9212999939918518\n"
     ]
    }
   ],
   "source": [
    "# obtention de l’accuracy sur les données de validation\n",
    "print(\"Pourcentage de bien classées :\", score[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On voit qu’on est au-delà de 92 % bien classés sur l’échantillon de validation.\n",
    "Dans notre cas, on a utilisé des images directement dans le package Keras."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Généralement, les images sont stockées dans des répertoires différents. Keras\n",
    "possède de nombreux outils pour charger ces images, on peut par exemple utiliser\n",
    "ImageDataGenerator :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 0 images belonging to 0 classes.\n",
      "Found 0 images belonging to 0 classes.\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "# on définit un générateur qui pourrait modifier nos images\n",
    "# (dans ce cas on passe en float 0 à 1)\n",
    "train_datagen = ImageDataGenerator(rescale=1./255)\n",
    "test_datagen = ImageDataGenerator(rescale=1./255)\n",
    "\n",
    "# on définit le dossier dans lequel se trouve les images\n",
    "train_generator = train_datagen.flow_from_directory('../data/',target_size=(150, 150),\n",
    "                                                    batch_size=32,class_mode='binary')\n",
    "# on définit le dossier dans lequel se trouve les images\n",
    "validation_generator = test_datagen.flow_from_directory('../data/',\n",
    "                                                        target_size=(150, 150),\n",
    "                                                        batch_size=32,\n",
    "                                                        class_mode='binary')"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "# on ajuste notre modèle de deep learning avec les images\n",
    "# de nos dossiers\n",
    "model.fit_generator(train_generator,epochs=50,\n",
    "                    validation_data=validation_generator)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Vous avez maintenant tous les outils pour vous lancer dans la construction de\n",
    "réseaux de deep learning avec Python."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.7.4 Aller plus loin : génération de features, transfer learning, RN, GAN …\n",
    "On charge facilement un modèle avec :"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.applications.resnet50 import ResNet50\n",
    "from tensorflow.keras.preprocessing import image\n",
    "from tensorflow.keras.applications.resnet50 import preprocess_input\n",
    "from tensorflow.keras.applications.resnet50 import decode_predictions\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# on utilise le modèle ResNet50\n",
    "model = ResNet50(weights='imagenet')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# on récupère une nouvelle image\n",
    "img_path = '../data/elephant.png'\n",
    "img = image.load_img(img_path, target_size=(224, 224))\n",
    "x = image.img_to_array(img)\n",
    "x = np.expand_dims(x, axis=0)\n",
    "x = preprocess_input(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 2s/step\n",
      "Prédit (classe/ %): [('n02504458', 'African_elephant', 0.6474553)]\n"
     ]
    }
   ],
   "source": [
    "# on applique le modèle\n",
    "preds = model.predict(x)\n",
    "print('Prédit (classe/ %):', decode_predictions(preds, top=1)[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On obtient donc un résultat à partir du modèle chargé. On pourra utiliser ce modèle\n",
    "et ses poids dans des couches intermédiaires de notre modèle."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Dans ce cas, on a chargé un modèle et on lui a fourni une photographie d’éléphant.\n",
    "Le modèle considère que c’est un éléphant d’Afrique à 87%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Les modèles de langages et le principe de l’attention\n",
    "\n",
    "Les modèles de langages (LLM) sont aujourd'hui centraux dans le cadre de l'IA. \n",
    "\n",
    "Python par l'intermédiaire du package transformers de Hugging-Face permet de charger et d'appliquer ce type de modèle. En voici un exemple rapide avec le modèle GPT2."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "What is python?\n",
      "\n",
      "Python is a programming language that allows you to write code that runs on top of other languages. It's a great way to learn about programming languages and how they can be used in your own projects. You can learn more about Python at http://pypi.python.org/wiki/Python.\n"
     ]
    }
   ],
   "source": [
    "# on importe les classes\n",
    "from transformers import GPT2LMHeadModel , GPT2Tokenizer \n",
    "# on charge le modèle de tokenizer\n",
    "tokenizer = GPT2Tokenizer.from_pretrained('gpt2') \n",
    "# on crée le modèle de langage à partir de GPT2\n",
    "model = GPT2LMHeadModel.from_pretrained('gpt2' , \n",
    "                        pad_token_id = tokenizer.eos_token_id)\n",
    "# on crée l'input\n",
    "question = 'What is python?' \n",
    "# on encode l'input grâce au tokenizer\n",
    "input_ids = tokenizer.encode(question, return_tensors='pt') \n",
    "# on génère l'output grâce au modèle\n",
    "output = model.generate(input_ids, \n",
    "                        max_length = 1000, \n",
    "                        num_beams = 5,\n",
    "                        no_repeat_ngram_size = 2,\n",
    "                        early_stopping = True\n",
    "                       )\n",
    "# on affiche la sortie\n",
    "print(tokenizer.decode(output[0], skip_special_tokens=True )) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "De nombreuses applications sont disponibles notamment en utilisant les modèles de Hugging Face et les API des différents modèles.\n",
    "\n",
    "Le développement de ces approches nécessaiterait un second ouvrage.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Nous avons effectué un panorama approfondi de l’utilisation de Python\n",
    "pour faire du machine learning, du deep learning et du traitement des\n",
    "données textuelles. Python ressort comme un langage adapté à toutes\n",
    "ces approches et un outil indispensable pour le data scientist."
   ]
  }
 ],
 "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": "3.12.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}