Skip to content

Instantly share code, notes, and snippets.

@harpiechoise

harpiechoise/Tutorial 2 Redes Convolucionales Keras Y Cifar-10.ipynb

Created January 15, 2018 01:05
Show Gist options
  • Save harpiechoise/009a4032695c94e95b8dc7beef3eef9d to your computer and use it in GitHub Desktop.
Save harpiechoise/009a4032695c94e95b8dc7beef3eef9d to your computer and use it in GitHub Desktop.
Download ZIP
Display the source blob
Display the rendered blob
Raw
Tutorial 2 Redes Convolucionales Keras Y Cifar-10.ipynb
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using TensorFlow backend.\n"
]
}
],
"source": [
"from __future__ import print_function\n",
"import numpy as np\n",
"from keras.datasets import cifar10\n",
"from keras.models import Sequential\n",
"from keras.layers import Conv2D,Flatten, Activation,MaxPooling2D,Dense\n",
"from keras.optimizers import RMSprop\n",
"from keras.utils import np_utils\n",
"from keras.layers import Dropout"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"IMG_CHANELS = 3 # Los colores de la imagen en este car RGB son 3 colores\n",
"IMG_ROWS = 32 # Largo de la imagen 32x32\n",
"IMG_COLS = 32 # Ancho de la imagen 32x32\n",
"\n",
"#Constantes\n",
"BATCH_SIZE = 128\n",
"NB_EPOCHS = 20\n",
"NB_CLASSES = 10\n",
"VERBOSE = 1\n",
"VALIDATION_SPLIT = 0.2\n",
"OPIMIZER = RMSprop()"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"#Dataset\n",
"(X_train,y_train),(X_test,y_test) = cifar10.load_data()\n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Formato del X_Train: (50000, 32, 32, 3)\n",
"Tamaño del X_Train 50000\n",
"Formato del X_test (10000, 32, 32, 3)\n",
"Tamaño del X_test 10000\n"
]
}
],
"source": [
"print(\"Formato del X_Train: \",X_train.shape)\n",
"print(\"Tamaño del X_Train\",X_train.shape[0])\n",
"print(\"Formato del X_test\",X_test.shape)\n",
"print(\"Tamaño del X_test\",X_test.shape[0])"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"#De numeros a one hot\n",
"Y_train = np_utils.to_categorical(y_train,NB_CLASSES)\n",
"Y_test = np_utils.to_categorical(y_test,NB_CLASSES)\n",
"#Normalizacion y etiquetado\n",
"X_train = X_train.astype('float32')\n",
"X_test = X_test.astype('float32')\n",
"X_train /= 255\n",
"X_test /= 255"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"_________________________________________________________________\n",
"Layer (type) Output Shape Param # \n",
"=================================================================\n",
"conv2d_1 (Conv2D) (None, 32, 32, 32) 896 \n",
"_________________________________________________________________\n",
"activation_1 (Activation) (None, 32, 32, 32) 0 \n",
"_________________________________________________________________\n",
"max_pooling2d_1 (MaxPooling2 (None, 16, 16, 32) 0 \n",
"_________________________________________________________________\n",
"dropout_1 (Dropout) (None, 16, 16, 32) 0 \n",
"_________________________________________________________________\n",
"flatten_1 (Flatten) (None, 8192) 0 \n",
"_________________________________________________________________\n",
"dense_1 (Dense) (None, 10) 81930 \n",
"_________________________________________________________________\n",
"activation_2 (Activation) (None, 10) 0 \n",
"=================================================================\n",
"Total params: 82,826\n",
"Trainable params: 82,826\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n"
]
}
],
"source": [
"#Red\n",
"model = Sequential()\n",
"model.add(Conv2D(32, kernel_size=3,padding='same',\n",
" input_shape=(IMG_ROWS,IMG_COLS,IMG_CHANELS)))\n",
"model.add(Activation('relu'))\n",
"model.add(MaxPooling2D(pool_size=(2,2)))\n",
"model.add(Dropout(.25))\n",
"model.add(Flatten())\n",
"model.add(Dense(NB_CLASSES))\n",
"model.add(Activation('softmax'))\n",
"model.summary()"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Train on 40000 samples, validate on 10000 samples\n",
"Epoch 1/20\n",
"40000/40000 [==============================] - 34s 845us/step - loss: 1.6893 - acc: 0.4082 - val_loss: 1.5426 - val_acc: 0.4617\n",
"Epoch 2/20\n",
"40000/40000 [==============================] - 31s 764us/step - loss: 1.4003 - acc: 0.5152 - val_loss: 1.3866 - val_acc: 0.5228\n",
"Epoch 3/20\n",
"40000/40000 [==============================] - 32s 806us/step - loss: 1.2929 - acc: 0.5518 - val_loss: 1.3567 - val_acc: 0.5334\n",
"Epoch 4/20\n",
"40000/40000 [==============================] - 33s 815us/step - loss: 1.2314 - acc: 0.5727 - val_loss: 1.3008 - val_acc: 0.5383\n",
"Epoch 5/20\n",
"40000/40000 [==============================] - 33s 837us/step - loss: 1.1876 - acc: 0.5898 - val_loss: 1.2161 - val_acc: 0.5788\n",
"Epoch 6/20\n",
"40000/40000 [==============================] - 33s 834us/step - loss: 1.1552 - acc: 0.5986 - val_loss: 1.2163 - val_acc: 0.5816\n",
"Epoch 7/20\n",
"40000/40000 [==============================] - 34s 840us/step - loss: 1.1254 - acc: 0.6101 - val_loss: 1.3196 - val_acc: 0.5438\n",
"Epoch 8/20\n",
"40000/40000 [==============================] - 35s 881us/step - loss: 1.1005 - acc: 0.6184 - val_loss: 1.1780 - val_acc: 0.5934\n",
"Epoch 9/20\n",
"40000/40000 [==============================] - 39s 980us/step - loss: 1.0804 - acc: 0.6282 - val_loss: 1.2367 - val_acc: 0.5700\n",
"Epoch 10/20\n",
"40000/40000 [==============================] - 38s 947us/step - loss: 1.0619 - acc: 0.6346 - val_loss: 1.2038 - val_acc: 0.5821\n",
"Epoch 11/20\n",
"40000/40000 [==============================] - 41s 1ms/step - loss: 1.0429 - acc: 0.6405 - val_loss: 1.2641 - val_acc: 0.5654\n",
"Epoch 12/20\n",
"40000/40000 [==============================] - 39s 978us/step - loss: 1.0321 - acc: 0.6427 - val_loss: 1.1748 - val_acc: 0.5902\n",
"Epoch 13/20\n",
"40000/40000 [==============================] - 40s 990us/step - loss: 1.0165 - acc: 0.6502 - val_loss: 1.2047 - val_acc: 0.5834\n",
"Epoch 14/20\n",
"40000/40000 [==============================] - 39s 978us/step - loss: 1.0023 - acc: 0.6542 - val_loss: 1.1386 - val_acc: 0.6053\n",
"Epoch 15/20\n",
"40000/40000 [==============================] - 39s 975us/step - loss: 0.9910 - acc: 0.6593 - val_loss: 1.0925 - val_acc: 0.6257\n",
"Epoch 16/20\n",
"40000/40000 [==============================] - 41s 1ms/step - loss: 0.9761 - acc: 0.6628 - val_loss: 1.1304 - val_acc: 0.6117\n",
"Epoch 17/20\n",
"40000/40000 [==============================] - 39s 974us/step - loss: 0.9669 - acc: 0.6653 - val_loss: 1.1060 - val_acc: 0.6246\n",
"Epoch 18/20\n",
"40000/40000 [==============================] - 41s 1ms/step - loss: 0.9552 - acc: 0.6712 - val_loss: 1.2358 - val_acc: 0.5839\n",
"Epoch 19/20\n",
"40000/40000 [==============================] - 39s 972us/step - loss: 0.9461 - acc: 0.6750 - val_loss: 1.1343 - val_acc: 0.6080\n",
"Epoch 20/20\n",
"40000/40000 [==============================] - 39s 982us/step - loss: 0.9352 - acc: 0.6777 - val_loss: 1.1561 - val_acc: 0.5994\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x7f9875fb2eb8>"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#Entrenar\n",
"model.compile(loss='categorical_crossentropy',\n",
" optimizer=OPIMIZER,\n",
" metrics=['accuracy'])\n",
"\n",
"model.fit(X_train,Y_train,batch_size=BATCH_SIZE,\n",
" epochs=NB_EPOCHS,\n",
" validation_split=VALIDATION_SPLIT,\n",
" verbose = VERBOSE)\n"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"10000/10000 [==============================] - 3s 321us/step\n",
"Puntaje: 1.14655178051\n",
"Presicion 0.5979\n"
]
}
],
"source": [
"#Evaluando\n",
"score = model.evaluate(X_test,Y_test,\n",
" batch_size=BATCH_SIZE,verbose=VERBOSE)\n",
"print(\"Puntaje: \",score[0])\n",
"print('Presicion',score[1])"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"\n",
"#Guardado de modelo\n",
"model.save_weights(\"cifar10_pesos.h5\",overwrite=True)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"_________________________________________________________________\n",
"Layer (type) Output Shape Param # \n",
"=================================================================\n",
"conv2d_2 (Conv2D) (None, 32, 32, 32) 896 \n",
"_________________________________________________________________\n",
"activation_3 (Activation) (None, 32, 32, 32) 0 \n",
"_________________________________________________________________\n",
"conv2d_3 (Conv2D) (None, 32, 32, 32) 9248 \n",
"_________________________________________________________________\n",
"activation_4 (Activation) (None, 32, 32, 32) 0 \n",
"_________________________________________________________________\n",
"max_pooling2d_2 (MaxPooling2 (None, 16, 16, 32) 0 \n",
"_________________________________________________________________\n",
"dropout_2 (Dropout) (None, 16, 16, 32) 0 \n",
"_________________________________________________________________\n",
"conv2d_4 (Conv2D) (None, 16, 16, 64) 18496 \n",
"_________________________________________________________________\n",
"activation_5 (Activation) (None, 16, 16, 64) 0 \n",
"_________________________________________________________________\n",
"conv2d_5 (Conv2D) (None, 14, 14, 64) 36928 \n",
"_________________________________________________________________\n",
"activation_6 (Activation) (None, 14, 14, 64) 0 \n",
"_________________________________________________________________\n",
"max_pooling2d_3 (MaxPooling2 (None, 7, 7, 64) 0 \n",
"_________________________________________________________________\n",
"dropout_3 (Dropout) (None, 7, 7, 64) 0 \n",
"_________________________________________________________________\n",
"flatten_2 (Flatten) (None, 3136) 0 \n",
"_________________________________________________________________\n",
"dense_2 (Dense) (None, 512) 1606144 \n",
"_________________________________________________________________\n",
"dropout_4 (Dropout) (None, 512) 0 \n",
"_________________________________________________________________\n",
"dense_3 (Dense) (None, 10) 5130 \n",
"_________________________________________________________________\n",
"activation_7 (Activation) (None, 10) 0 \n",
"=================================================================\n",
"Total params: 1,676,842\n",
"Trainable params: 1,676,842\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n"
]
}
],
"source": [
"#Modelo 2 mejorado\n",
"model2 = Sequential()\n",
"model2.add(Conv2D(32,kernel_size=3,padding='same',\n",
" input_shape=(IMG_ROWS,IMG_COLS,IMG_CHANELS)))\n",
"model2.add(Activation('relu'))\n",
"model2.add(Conv2D(32,kernel_size=3,padding='same'))\n",
"model2.add(Activation('relu'))\n",
"model2.add(MaxPooling2D(pool_size=(2,2)))\n",
"model2.add(Dropout(.25))\n",
"model2.add(Conv2D(64,kernel_size=3,padding='same'))\n",
"model2.add(Activation('relu'))\n",
"model2.add(Conv2D(64,kernel_size=3))\n",
"model2.add(Activation('relu'))\n",
"model2.add(MaxPooling2D(pool_size=(2,2)))\n",
"model2.add(Dropout(.25))\n",
"model2.add(Flatten())\n",
"model2.add(Dense(512))\n",
"model2.add(Dropout(.5))\n",
"model2.add(Dense(NB_CLASSES))\n",
"model2.add(Activation('softmax'))\n",
"model2.summary()\n",
"#Entrenar\n",
"model2.compile(loss='categorical_crossentropy',\n",
" optimizer=OPIMIZER,\n",
" metrics=['accuracy'])"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Train on 40000 samples, validate on 10000 samples\n",
"Epoch 1/20\n",
"40000/40000 [==============================] - 304s 8ms/step - loss: 1.6460 - acc: 0.4048 - val_loss: 1.2532 - val_acc: 0.5528\n",
"Epoch 2/20\n",
"40000/40000 [==============================] - 307s 8ms/step - loss: 1.1826 - acc: 0.5829 - val_loss: 0.9569 - val_acc: 0.6681\n",
"Epoch 3/20\n",
"40000/40000 [==============================] - 308s 8ms/step - loss: 1.0073 - acc: 0.6502 - val_loss: 1.0642 - val_acc: 0.6326\n",
"Epoch 4/20\n",
"40000/40000 [==============================] - 303s 8ms/step - loss: 0.9107 - acc: 0.6830 - val_loss: 0.7950 - val_acc: 0.7166\n",
"Epoch 5/20\n",
"40000/40000 [==============================] - 298s 7ms/step - loss: 0.8442 - acc: 0.7063 - val_loss: 0.7784 - val_acc: 0.7213\n",
"Epoch 6/20\n",
"40000/40000 [==============================] - 303s 8ms/step - loss: 0.7967 - acc: 0.7242 - val_loss: 0.7338 - val_acc: 0.7451\n",
"Epoch 7/20\n",
"40000/40000 [==============================] - 315s 8ms/step - loss: 0.7647 - acc: 0.7328 - val_loss: 0.7674 - val_acc: 0.7343\n",
"Epoch 8/20\n",
"40000/40000 [==============================] - 312s 8ms/step - loss: 0.7390 - acc: 0.7440 - val_loss: 0.7710 - val_acc: 0.7352\n",
"Epoch 9/20\n",
"40000/40000 [==============================] - 316s 8ms/step - loss: 0.7103 - acc: 0.7530 - val_loss: 0.8495 - val_acc: 0.7170\n",
"Epoch 10/20\n",
"40000/40000 [==============================] - 310s 8ms/step - loss: 0.6906 - acc: 0.7598 - val_loss: 0.7262 - val_acc: 0.7538\n",
"Epoch 11/20\n",
"40000/40000 [==============================] - 308s 8ms/step - loss: 0.6642 - acc: 0.7689 - val_loss: 0.7318 - val_acc: 0.7527\n",
"Epoch 12/20\n",
"40000/40000 [==============================] - 306s 8ms/step - loss: 0.6532 - acc: 0.7751 - val_loss: 0.7413 - val_acc: 0.7478\n",
"Epoch 13/20\n",
"40000/40000 [==============================] - 310s 8ms/step - loss: 0.6425 - acc: 0.7772 - val_loss: 0.6976 - val_acc: 0.7621\n",
"Epoch 14/20\n",
"40000/40000 [==============================] - 310s 8ms/step - loss: 0.6311 - acc: 0.7813 - val_loss: 0.7028 - val_acc: 0.7649\n",
"Epoch 15/20\n",
"40000/40000 [==============================] - 310s 8ms/step - loss: 0.6109 - acc: 0.7874 - val_loss: 0.7126 - val_acc: 0.7628\n",
"Epoch 16/20\n",
"40000/40000 [==============================] - 310s 8ms/step - loss: 0.6061 - acc: 0.7907 - val_loss: 0.6751 - val_acc: 0.7751\n",
"Epoch 17/20\n",
"40000/40000 [==============================] - 305s 8ms/step - loss: 0.5889 - acc: 0.7962 - val_loss: 0.7055 - val_acc: 0.7604\n",
"Epoch 18/20\n",
"40000/40000 [==============================] - 301s 8ms/step - loss: 0.5830 - acc: 0.7972 - val_loss: 0.7047 - val_acc: 0.7707\n",
"Epoch 19/20\n",
"40000/40000 [==============================] - 301s 8ms/step - loss: 0.5762 - acc: 0.7982 - val_loss: 0.6651 - val_acc: 0.7773\n",
"Epoch 20/20\n",
"40000/40000 [==============================] - 299s 7ms/step - loss: 0.5630 - acc: 0.8051 - val_loss: 0.7770 - val_acc: 0.7527\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x7f9875fb2c18>"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"\n",
"\n",
"model2.fit(X_train,Y_train,batch_size=64,\n",
" epochs=NB_EPOCHS,\n",
" validation_split=VALIDATION_SPLIT,\n",
" verbose = VERBOSE)"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [],
"source": [
"model2.save_weights('cifar 10 pesos modelo 2.h5',overwrite=True)"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import scipy.misc\n",
"from keras.models import model_from_json\n",
"from keras.optimizers import SGD"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[3]\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/home/harpie/anaconda/envs/keras/lib/python3.5/site-packages/ipykernel_launcher.py:2: DeprecationWarning: `imread` is deprecated!\n",
"`imread` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.\n",
"Use ``imageio.imread`` instead.\n",
" \n",
"/home/harpie/anaconda/envs/keras/lib/python3.5/site-packages/ipykernel_launcher.py:2: DeprecationWarning: `imresize` is deprecated!\n",
"`imresize` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.\n",
"Use ``skimage.transform.resize`` instead.\n",
" \n"
]
}
],
"source": [
"img_names = [\"imagen/gato.jpg\"]\n",
"imgs = [np.transpose(scipy.misc.imresize(scipy.misc.imread(x),(32,32)),(1,0,2)).astype(\"float32\") for x in img_names]\n",
"imgs = np.array(imgs) / 255\n",
"# predicciones\n",
"pred = model2.predict_classes(imgs)\n",
"print(pred)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"https://www.cs.toronto.edu/~kriz/cifar.html \n",
"\n",
"# segun cifar-10 tenemos 10 categorias evaluadas del 0 al 9 \n",
"# 1. Aeroplanos\n",
"# 2. Automoviles\n",
"# 3. Pajaros\n",
"# 4. Gatos\n",
"# 5. Benados\n",
"# 6. Perros\n",
"# 7. Ranas \n",
"# 8. Caballos\n",
"# 9. Obejas\n",
"# 10. Camiones\n",
"\n",
"# Si seguimos la logica de empezar por el 0 el 3 serian gatos y por la imagen mostrada es correcta la prediccion"
]
}
],
"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": "3.5.2"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment