{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "personal-marshall",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np \n",
    "import os\n",
    "import pandas as pd \n",
    "from scipy.io import wavfile\n",
    "\n",
    "import librosa\n",
    "from tqdm import tqdm\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "from tensorflow.keras import regularizers, activations\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Dropout, Flatten, Activation, Conv2D, MaxPooling2D, GlobalAveragePooling2D\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "from datetime import datetime \n",
    "\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "import cv2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "intimate-property",
   "metadata": {},
   "outputs": [],
   "source": [
    "us8k_df = pd.read_pickle(\"us8k_df.pkl\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "closed-abraham",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = us8k_df.drop(['fold'],axis=1)\n",
    "X = np.stack(df.melspectrogram.to_numpy())\n",
    "X_dim = (128,128,1)\n",
    "X = X.reshape(X.shape[0], *X_dim)\n",
    "Y = np.array(df['label'])\n",
    "Y = to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "controlling-overhead",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((8732, 10), (8732, 128, 128, 1))"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y.shape, X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "dying-cooking",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_new = np.zeros((8732,128,128,3))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "increasing-annotation",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(8732, 128, 128, 3)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_new.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "rubber-hanging",
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(len(X)):\n",
    "    X_new[i]=cv2.cvtColor(X[i], cv2.COLOR_GRAY2RGB)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "after-driving",
   "metadata": {},
   "outputs": [],
   "source": [
    "X=X_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "illegal-partner",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, Y_train, Y_test = train_test_split(X,Y,test_size=0.3,shuffle=True,stratify = Y)\n",
    "X_val, X_test, Y_val,Y_test = train_test_split(X_test,Y_test,test_size=0.5,shuffle=True,stratify = Y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "french-assembly",
   "metadata": {},
   "source": [
    "MOdèle utilisant mobile net "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "geographic-diving",
   "metadata": {},
   "outputs": [],
   "source": [
    "preprocess_input = tf.keras.applications.mobilenet_v2.preprocess_input\n",
    "rescale = tf.keras.layers.experimental.preprocessing.Rescaling(1./127.5, offset= -1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "pacific-registration",
   "metadata": {},
   "outputs": [],
   "source": [
    "IMG_SHAPE = (128,128,3)\n",
    "\n",
    "base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,\n",
    "                                               include_top=False,\n",
    "                                               weights='imagenet')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "included-mozambique",
   "metadata": {},
   "outputs": [],
   "source": [
    "prediction_layer = tf.keras.layers.Dense(10)\n",
    "global_average_layer = tf.keras.layers.GlobalAveragePooling2D()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "palestinian-crowd",
   "metadata": {},
   "outputs": [],
   "source": [
    "# model : MobileNet puis du dropout et une couche dense pour la prédiction\n",
    "inputs = tf.keras.Input(shape=IMG_SHAPE)\n",
    "x = preprocess_input(inputs)\n",
    "x = rescale(x)\n",
    "x = base_model(x, training=False)\n",
    "x = global_average_layer(x)\n",
    "x = tf.keras.layers.Dropout(0.2)(x)\n",
    "outputs = prediction_layer(x)\n",
    "model = tf.keras.Model(inputs, outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "simple-rebecca",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         [(None, 128, 128, 3)]     0         \n",
      "_________________________________________________________________\n",
      "tf.math.truediv (TFOpLambda) (None, 128, 128, 3)       0         \n",
      "_________________________________________________________________\n",
      "tf.math.subtract (TFOpLambda (None, 128, 128, 3)       0         \n",
      "_________________________________________________________________\n",
      "rescaling (Rescaling)        (None, 128, 128, 3)       0         \n",
      "_________________________________________________________________\n",
      "mobilenetv2_1.00_128 (Functi (None, 4, 4, 1280)        2257984   \n",
      "_________________________________________________________________\n",
      "global_average_pooling2d (Gl (None, 1280)              0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 1280)              0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 10)                12810     \n",
      "=================================================================\n",
      "Total params: 2,270,794\n",
      "Trainable params: 2,236,682\n",
      "Non-trainable params: 34,112\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "pacific-correction",
   "metadata": {},
   "outputs": [],
   "source": [
    "base_learning_rate = 0.001\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(lr=base_learning_rate),\n",
    "              loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "durable-straight",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "41/41 [==============================] - 5s 99ms/step - loss: 0.9276 - accuracy: 0.1113\n"
     ]
    }
   ],
   "source": [
    "initial_epochs = 10\n",
    "num_batch_size = 32\n",
    "loss0, accuracy0 = model.evaluate(X_val,Y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "victorian-hawaiian",
   "metadata": {},
   "outputs": [],
   "source": [
    "log_dir = \"logs/fit/\" + datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n",
    "tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir = log_dir, histogram_freq = 1)\n",
    "save_best = tf.keras.callbacks.ModelCheckpoint(filepath = \"logs/checkpoints/\", save_weights_only = True,\n",
    "                                               monitor = \"val_accuracy\", mode = \"max\", save_best_only = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "adolescent-prerequisite",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "63/63 [==============================] - 37s 552ms/step - loss: 0.4256 - accuracy: 0.1145 - val_loss: 0.3422 - val_accuracy: 0.1133\n",
      "Epoch 2/10\n",
      " 4/63 [>.............................] - ETA: 39s - loss: 0.3885 - accuracy: 0.0859"
     ]
    }
   ],
   "source": [
    "model_fit = model.fit(X_train[:2000],Y_train[0:2000], epochs=initial_epochs,validation_data=(X_val[0:300],Y_val[0:300]),batch_size=num_batch_size,callbacks = [tensorboard_callback, save_best])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "civic-trinity",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.2 64-bit",
   "language": "python",
   "name": "python38264bit553a082d8987410db038174ed8b7851d"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}