E_Fine_Tuning_VGG19_1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Entreno de un ConvNet VGG19 preentrenado con Imagenet y\n",
    "entrenando conjunto de datos pequeño y FINE TUNING\n",
    "Conjunto de datos de 4 clases\n",
    "Utilizaremos un conjunto de datos reducido:\n",
    "- Conjunto de entrenamiento: 1.000 muestras x clase \n",
    "- Conjunto de validación: 300 muestras x clase\n",
    "- Conjunto de test: 300 muestras x clase\n",
    "El conjunto de datos: balanced_dataset\n",
    "contiene tres carpetas, train, validation y test, cada una con 4 clases"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.5/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from  keras  import  applications\n",
    "from  keras.preprocessing.image  import  ImageDataGenerator \n",
    "from keras.optimizers import SGD\n",
    "from  keras.applications.vgg19 import VGG19 \n",
    "from  keras.models  import  Sequential,  Model\n",
    "from  keras.layers  import  Dropout,  Flatten,  Dense, Input \n",
    "top_model_weights_path  =  'VGG19_FE_model_3.h5'\n",
    "\n",
    "img_width = 224\n",
    "img_height = 224\n",
    "\n",
    "nb_epochs = 60\n",
    "batch_size  =  64  #  reduce  to  5  in  order  to  be  able  to  compute  the  calcs\n",
    "\n",
    "train_dir  =  \"balanced_dataset/train\" \n",
    "validation_dir = \"balanced_dataset/validation\" \n",
    "test_dir  =  \"balanced_dataset/test\"\n",
    "\n",
    "NB_VGG19_FREEZE_LAYERS = 4\n",
    "nb_train = 4000\n",
    "nb_validation = 1200\n",
    "nb_test = 1200\n",
    "nb_classes = 4\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "input_tensor = Input(shape=(img_width, img_height, 3))  #  another  way  to  shape  the  input\n",
    "\n",
    "base_model = applications.VGG19(\n",
    "    weights='imagenet',  \n",
    "    include_top=False, \n",
    "    input_tensor=input_tensor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.5/dist-packages/ipykernel_launcher.py:12: UserWarning: Update your `Model` call to the Keras 2 API: `Model(outputs=Tensor(\"se..., inputs=Tensor(\"in...)`\n",
      "  if sys.path[0] == '':\n"
     ]
    }
   ],
   "source": [
    "# top model\n",
    "top_model = Sequential() \n",
    "top_model.add(Flatten(input_shape=base_model.output_shape[1:])) \n",
    "top_model.add(Dense(256,  activation='relu')) \n",
    "top_model.add(Dropout(0.5))\n",
    "top_model.add(Dense(nb_classes,  activation='softmax'))\n",
    "\n",
    "# base model has its weights, now we load the weights on the top layer \n",
    "top_model.load_weights(\"VGG/VGG19_FE_model_3.h5\")\n",
    "\n",
    "# we join base and top it has to be updated to api2\n",
    "model_total  =  Model(input = base_model.input,  output = top_model(base_model.output))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 224, 224, 3)       0         \n",
      "_________________________________________________________________\n",
      "block1_conv1 (Conv2D)        (None, 224, 224, 64)      1792      \n",
      "_________________________________________________________________\n",
      "block1_conv2 (Conv2D)        (None, 224, 224, 64)      36928     \n",
      "_________________________________________________________________\n",
      "block1_pool (MaxPooling2D)   (None, 112, 112, 64)      0         \n",
      "_________________________________________________________________\n",
      "block2_conv1 (Conv2D)        (None, 112, 112, 128)     73856     \n",
      "_________________________________________________________________\n",
      "block2_conv2 (Conv2D)        (None, 112, 112, 128)     147584    \n",
      "_________________________________________________________________\n",
      "block2_pool (MaxPooling2D)   (None, 56, 56, 128)       0         \n",
      "_________________________________________________________________\n",
      "block3_conv1 (Conv2D)        (None, 56, 56, 256)       295168    \n",
      "_________________________________________________________________\n",
      "block3_conv2 (Conv2D)        (None, 56, 56, 256)       590080    \n",
      "_________________________________________________________________\n",
      "block3_conv3 (Conv2D)        (None, 56, 56, 256)       590080    \n",
      "_________________________________________________________________\n",
      "block3_conv4 (Conv2D)        (None, 56, 56, 256)       590080    \n",
      "_________________________________________________________________\n",
      "block3_pool (MaxPooling2D)   (None, 28, 28, 256)       0         \n",
      "_________________________________________________________________\n",
      "block4_conv1 (Conv2D)        (None, 28, 28, 512)       1180160   \n",
      "_________________________________________________________________\n",
      "block4_conv2 (Conv2D)        (None, 28, 28, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block4_conv3 (Conv2D)        (None, 28, 28, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block4_conv4 (Conv2D)        (None, 28, 28, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block4_pool (MaxPooling2D)   (None, 14, 14, 512)       0         \n",
      "_________________________________________________________________\n",
      "block5_conv1 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block5_conv2 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block5_conv3 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block5_conv4 (Conv2D)        (None, 14, 14, 512)       2359808   \n",
      "_________________________________________________________________\n",
      "block5_pool (MaxPooling2D)   (None, 7, 7, 512)         0         \n",
      "_________________________________________________________________\n",
      "sequential_1 (Sequential)    (None, 4)                 6423812   \n",
      "=================================================================\n",
      "Total params: 26,448,196\n",
      "Trainable params: 26,448,196\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_total.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Freeze the layers except the last 4 layers\n",
    "for layer in base_model.layers[:-NB_VGG19_FREEZE_LAYERS]:\n",
    "    layer.trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<keras.engine.topology.InputLayer object at 0x7f1e45d5b390> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d5b358> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d5b4e0> False\n",
      "<keras.layers.pooling.MaxPooling2D object at 0x7f1e45d5bf98> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d78780> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d8ee48> False\n",
      "<keras.layers.pooling.MaxPooling2D object at 0x7f1e45d30d30> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d415f8> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d51e80> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45d04240> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45cf49b0> False\n",
      "<keras.layers.pooling.MaxPooling2D object at 0x7f1e45d15630> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45cc7630> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45cc7f60> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45c69fd0> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45c7aa58> False\n",
      "<keras.layers.pooling.MaxPooling2D object at 0x7f1e45c1d400> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45c2e470> False\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e45c507f0> True\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e459c5eb8> True\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1e459d6470> True\n",
      "<keras.layers.pooling.MaxPooling2D object at 0x7f1e45988f28> True\n"
     ]
    }
   ],
   "source": [
    "# Check the trainable status of the individual layers\n",
    "for layer in base_model.layers:\n",
    "    print(layer, layer.trainable)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 4000 images belonging to 4 classes.\n",
      "Found 1200 images belonging to 4 classes.\n",
      "Found 1200 images belonging to 4 classes.\n"
     ]
    }
   ],
   "source": [
    "#PREPROCESSING + DATA AUGMENTATION\n",
    "train_datagen = ImageDataGenerator(\n",
    "    rescale=1./255, \n",
    "    rotation_range=40, \n",
    "    width_shift_range=0.2, \n",
    "    height_shift_range=0.2, \n",
    "    zoom_range=0.2, \n",
    "    horizontal_flip=True, \n",
    "    vertical_flip  =  True, \n",
    "    fill_mode='nearest')\n",
    "\n",
    "train_generator = train_datagen.flow_from_directory( \n",
    "    train_dir, \n",
    "    target_size=(img_width,img_height), \n",
    "    batch_size=batch_size, \n",
    "    class_mode='categorical')\n",
    "\n",
    "test_datagen  =  ImageDataGenerator(rescale=1. / 255)  #  not  in  the  val  data \n",
    "\n",
    "validation_generator = test_datagen.flow_from_directory(\n",
    "    validation_dir,\n",
    "    target_size=(img_height, img_width), \n",
    "    batch_size=batch_size, \n",
    "    class_mode='categorical')\n",
    "\n",
    "test_generator = test_datagen.flow_from_directory(\n",
    "    test_dir,\n",
    "    target_size=(img_height, img_width), \n",
    "    batch_size=batch_size, \n",
    "    class_mode='categorical')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Compile model\n",
    "model_total.compile(optimizer=SGD(lr=1e-4,  momentum=0.9), \n",
    "                    loss='categorical_crossentropy',  \n",
    "                    metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.5/dist-packages/ipykernel_launcher.py:8: UserWarning: Update your `fit_generator` call to the Keras 2 API: `fit_generator(<keras.pre..., steps_per_epoch=62, verbose=1, validation_steps=1200, validation_data=<keras.pre..., epochs=60)`\n",
      "  \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/60\n",
      "62/62 [==============================] - 628s 10s/step - loss: 0.8835 - acc: 0.6497 - val_loss: 0.6354 - val_acc: 0.7374\n",
      "Epoch 2/60\n",
      "62/62 [==============================] - 626s 10s/step - loss: 0.7005 - acc: 0.7177 - val_loss: 0.5066 - val_acc: 0.7934\n",
      "Epoch 3/60\n",
      "62/62 [==============================] - 624s 10s/step - loss: 0.6826 - acc: 0.7359 - val_loss: 0.4764 - val_acc: 0.8209\n",
      "Epoch 4/60\n",
      "62/62 [==============================] - 628s 10s/step - loss: 0.6275 - acc: 0.7510 - val_loss: 0.3997 - val_acc: 0.8351\n",
      "Epoch 5/60\n",
      "62/62 [==============================] - 625s 10s/step - loss: 0.5868 - acc: 0.7666 - val_loss: 0.4999 - val_acc: 0.7924\n",
      "Epoch 6/60\n",
      "62/62 [==============================] - 628s 10s/step - loss: 0.5735 - acc: 0.7752 - val_loss: 0.3838 - val_acc: 0.8451\n",
      "Epoch 7/60\n",
      "62/62 [==============================] - 624s 10s/step - loss: 0.5584 - acc: 0.7818 - val_loss: 0.3795 - val_acc: 0.8458\n",
      "Epoch 8/60\n",
      "62/62 [==============================] - 624s 10s/step - loss: 0.5088 - acc: 0.8004 - val_loss: 0.3991 - val_acc: 0.8409\n",
      "Epoch 9/60\n",
      "62/62 [==============================] - 624s 10s/step - loss: 0.5275 - acc: 0.7979 - val_loss: 0.3691 - val_acc: 0.8567\n",
      "Epoch 10/60\n",
      "62/62 [==============================] - 625s 10s/step - loss: 0.5248 - acc: 0.7906 - val_loss: 0.3917 - val_acc: 0.8483\n",
      "Epoch 11/60\n",
      "62/62 [==============================] - 621s 10s/step - loss: 0.5007 - acc: 0.8097 - val_loss: 0.3807 - val_acc: 0.8442\n",
      "Epoch 12/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.4629 - acc: 0.8266 - val_loss: 0.3684 - val_acc: 0.8607\n",
      "Epoch 13/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.4708 - acc: 0.8158 - val_loss: 0.3525 - val_acc: 0.8567\n",
      "Epoch 14/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.4748 - acc: 0.8206 - val_loss: 0.4010 - val_acc: 0.8376\n",
      "Epoch 15/60\n",
      "62/62 [==============================] - 620s 10s/step - loss: 0.4746 - acc: 0.8175 - val_loss: 0.3592 - val_acc: 0.8600\n",
      "Epoch 16/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.4559 - acc: 0.8254 - val_loss: 0.3289 - val_acc: 0.8585\n",
      "Epoch 17/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.4361 - acc: 0.8324 - val_loss: 0.4215 - val_acc: 0.8282\n",
      "Epoch 18/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.4195 - acc: 0.8390 - val_loss: 0.3377 - val_acc: 0.8667\n",
      "Epoch 19/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.4260 - acc: 0.8359 - val_loss: 0.3408 - val_acc: 0.8684\n",
      "Epoch 20/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.4300 - acc: 0.8309 - val_loss: 0.3329 - val_acc: 0.8649\n",
      "Epoch 21/60\n",
      "62/62 [==============================] - 620s 10s/step - loss: 0.4174 - acc: 0.8463 - val_loss: 0.4126 - val_acc: 0.8341\n",
      "Epoch 22/60\n",
      "62/62 [==============================] - 614s 10s/step - loss: 0.4118 - acc: 0.8405 - val_loss: 0.3567 - val_acc: 0.8608\n",
      "Epoch 23/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.4011 - acc: 0.8478 - val_loss: 0.3179 - val_acc: 0.8750\n",
      "Epoch 24/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.3826 - acc: 0.8543 - val_loss: 0.3415 - val_acc: 0.8650\n",
      "Epoch 25/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.3816 - acc: 0.8541 - val_loss: 0.3315 - val_acc: 0.8707\n",
      "Epoch 26/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.3923 - acc: 0.8495 - val_loss: 0.4144 - val_acc: 0.8541\n",
      "Epoch 27/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.4157 - acc: 0.8443 - val_loss: 0.3388 - val_acc: 0.8675\n",
      "Epoch 28/60\n",
      "62/62 [==============================] - 620s 10s/step - loss: 0.3895 - acc: 0.8551 - val_loss: 0.2914 - val_acc: 0.8825\n",
      "Epoch 29/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.3683 - acc: 0.8561 - val_loss: 0.2860 - val_acc: 0.8825\n",
      "Epoch 31/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3697 - acc: 0.8629 - val_loss: 0.2965 - val_acc: 0.8817\n",
      "Epoch 32/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3597 - acc: 0.8689 - val_loss: 0.2656 - val_acc: 0.8958\n",
      "Epoch 33/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3506 - acc: 0.8664 - val_loss: 0.2809 - val_acc: 0.8925\n",
      "Epoch 34/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.3751 - acc: 0.8589 - val_loss: 0.2783 - val_acc: 0.8942\n",
      "Epoch 35/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.3525 - acc: 0.8659 - val_loss: 0.3096 - val_acc: 0.8809\n",
      "Epoch 36/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3347 - acc: 0.8805 - val_loss: 0.3245 - val_acc: 0.8801\n",
      "Epoch 37/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3510 - acc: 0.8705 - val_loss: 0.3699 - val_acc: 0.8541\n",
      "Epoch 38/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3399 - acc: 0.8727 - val_loss: 0.2537 - val_acc: 0.8967\n",
      "Epoch 39/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.3342 - acc: 0.8785 - val_loss: 0.2796 - val_acc: 0.8909\n",
      "Epoch 40/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3156 - acc: 0.8861 - val_loss: 0.3001 - val_acc: 0.8850\n",
      "Epoch 41/60\n",
      "62/62 [==============================] - 613s 10s/step - loss: 0.3173 - acc: 0.8816 - val_loss: 0.2425 - val_acc: 0.9033\n",
      "Epoch 42/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3289 - acc: 0.8783 - val_loss: 0.3098 - val_acc: 0.8816\n",
      "Epoch 43/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.3405 - acc: 0.8737 - val_loss: 0.3168 - val_acc: 0.8725\n",
      "Epoch 44/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3432 - acc: 0.8735 - val_loss: 0.3578 - val_acc: 0.8600\n",
      "Epoch 45/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3305 - acc: 0.8758 - val_loss: 0.2726 - val_acc: 0.8942\n",
      "Epoch 46/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3228 - acc: 0.8760 - val_loss: 0.2798 - val_acc: 0.8925\n",
      "Epoch 47/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.3113 - acc: 0.8800 - val_loss: 0.2853 - val_acc: 0.8942\n",
      "Epoch 48/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.3202 - acc: 0.8800 - val_loss: 0.3432 - val_acc: 0.8676\n",
      "Epoch 49/60\n",
      "62/62 [==============================] - 619s 10s/step - loss: 0.3165 - acc: 0.8841 - val_loss: 0.3774 - val_acc: 0.8608\n",
      "Epoch 50/60\n",
      "62/62 [==============================] - 613s 10s/step - loss: 0.2984 - acc: 0.8926 - val_loss: 0.2833 - val_acc: 0.8899\n",
      "Epoch 51/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.3038 - acc: 0.8904 - val_loss: 0.2612 - val_acc: 0.9000\n",
      "Epoch 52/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.3029 - acc: 0.8876 - val_loss: 0.2387 - val_acc: 0.9101\n",
      "Epoch 53/60\n",
      "62/62 [==============================] - 616s 10s/step - loss: 0.2831 - acc: 0.8952 - val_loss: 0.2572 - val_acc: 0.9083\n",
      "Epoch 54/60\n",
      "62/62 [==============================] - 614s 10s/step - loss: 0.3061 - acc: 0.8884 - val_loss: 0.3213 - val_acc: 0.8882\n",
      "Epoch 55/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.3031 - acc: 0.8891 - val_loss: 0.2245 - val_acc: 0.9117\n",
      "Epoch 56/60\n",
      "62/62 [==============================] - 618s 10s/step - loss: 0.3045 - acc: 0.8886 - val_loss: 0.2306 - val_acc: 0.9126\n",
      "Epoch 57/60\n",
      "62/62 [==============================] - 613s 10s/step - loss: 0.3039 - acc: 0.8891 - val_loss: 0.3330 - val_acc: 0.8809\n",
      "Epoch 58/60\n",
      "62/62 [==============================] - 615s 10s/step - loss: 0.2817 - acc: 0.8899 - val_loss: 0.2306 - val_acc: 0.9075\n",
      "Epoch 59/60\n",
      "62/62 [==============================] - 620s 10s/step - loss: 0.3055 - acc: 0.8843 - val_loss: 0.2771 - val_acc: 0.8875\n",
      "Epoch 60/60\n",
      "62/62 [==============================] - 617s 10s/step - loss: 0.2861 - acc: 0.8914 - val_loss: 0.2586 - val_acc: 0.8883\n"
     ]
    }
   ],
   "source": [
    "#Fit model\n",
    "history = model_total.fit_generator( \n",
    "    train_generator, \n",
    "    samples_per_epoch=nb_train, \n",
    "    epochs=nb_epochs,\n",
    "    verbose  =  1, \n",
    "    validation_data=validation_generator, \n",
    "    validation_steps=nb_validation)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model_total.save_weights('VGG/VGG19_FT_model_1.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def print_graphics (history):\n",
    "    import  matplotlib.pyplot  as  plt\n",
    "    #Mostramos otro tipo de grafico\n",
    "    acc = history.history['acc']\n",
    "    val_acc = history.history['val_acc']\n",
    "    loss = history.history['loss']\n",
    "    val_loss = history.history['val_loss']\n",
    "\n",
    "    epochs = range(1, len(acc)+ 1)\n",
    "\n",
    "    plt.plot(epochs, acc, 'bo', label='Training acc')\n",
    "    plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
    "    #plt.tittle('Trainning and validation accuracy')\n",
    "    plt.legend()\n",
    "\n",
    "    plt.figure()\n",
    "    plt.plot(epochs, loss, 'bo', label='Training loss')\n",
    "    plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "    #plt.tittle('Trainning and validation loss')\n",
    "    plt.legend()\n",
    "\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1e3e63c198>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1e438e39e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print_graphics (history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.272254011631012, 0.899166665871938]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_total.evaluate_generator(test_generator)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}