Implementation of Convolutional Neural Network using Keras library.
%matplotlib inline
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image
import tensorflow as tf
plt.rcParams['figure.figsize'] = (20, 10)
cifar10 = tf.keras.datasets.cifar10
[X_train, Y_train],[X_test, Y_test] = cifar10.load_data()
# Data preparation
X_train, Y_train = X_train/255, Y_train
X_test, Y_test = X_test/255, Y_test
# Data augmentation [flip x]
X_train = np.vstack([X_train, X_train[:,:,::-1,:]])
Y_train = np.vstack([Y_train, Y_train])
X_test = np.vstack([X_test, X_test[:,:,::-1,:]])
Y_test = np.vstack([Y_test, Y_test])
# one hot for multi-class classification
# MNIST = 10 classes [0-9]
Y_train_oh = np.zeros((Y_train.size, 10))
Y_train_oh[np.arange(Y_train.size), Y_train.T] = 1
Y_test_oh = np.zeros((Y_test.size, 10))
Y_test_oh[np.arange(Y_test.size), Y_test.T] = 1
print('X_train:', X_train.shape)
print('Y_train:', Y_train.shape)
print('Y_train_oh:', Y_train_oh.shape)
print('X_test:', X_test.shape)
print('Y_test:', Y_test.shape)
print('Y_test_oh:', Y_test_oh.shape)
X_train: (100000, 32, 32, 3) Y_train: (100000, 1) Y_train_oh: (100000, 10) X_test: (20000, 32, 32, 3) Y_test: (20000, 1) Y_test_oh: (20000, 10)
# Label categories
objects = ['Plane', 'Car', 'Bird', 'Cat', 'Deer', 'Dog', 'Frog', 'Horse', 'Ship', 'Truck']
# Get dimensions
N1, N2, C = X_train[0].shape
fig, AX = plt.subplots(3, 6, sharex=True, sharey=True)
np.random.seed(1234)
for ax in AX.ravel():
rindex = np.random.randint(Y_train.size)
ax.imshow(X_train[rindex])
# title label + one-hot
title = f'{objects[Y_train[rindex][0]]} {Y_train[rindex]} :: '
title += ''.join([str(int(e)) for e in Y_train_oh[rindex]])
ax.set_title(title)
plt.grid(False)
# Model
model = tf.keras.Sequential([
tf.keras.layers.Input(X_train[0].shape),
tf.keras.layers.Conv2D(32, 3, activation='relu'),
tf.keras.layers.Conv2D(64, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(pool_size=2),
tf.keras.layers.Dropout(0.25),
tf.keras.layers.Conv2D(128, 3, activation='relu'),
tf.keras.layers.Conv2D(256, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(pool_size=2),
tf.keras.layers.Dropout(0.25),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(N1*N2*C, activation='relu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(10, activation='softmax')
])
model.summary()
Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d (Conv2D) (None, 30, 30, 32) 896 _________________________________________________________________ conv2d_1 (Conv2D) (None, 28, 28, 64) 18496 _________________________________________________________________ max_pooling2d (MaxPooling2D) (None, 14, 14, 64) 0 _________________________________________________________________ dropout (Dropout) (None, 14, 14, 64) 0 _________________________________________________________________ conv2d_2 (Conv2D) (None, 12, 12, 128) 73856 _________________________________________________________________ conv2d_3 (Conv2D) (None, 10, 10, 256) 295168 _________________________________________________________________ max_pooling2d_1 (MaxPooling2 (None, 5, 5, 256) 0 _________________________________________________________________ dropout_1 (Dropout) (None, 5, 5, 256) 0 _________________________________________________________________ flatten (Flatten) (None, 6400) 0 _________________________________________________________________ dense (Dense) (None, 3072) 19663872 _________________________________________________________________ dropout_2 (Dropout) (None, 3072) 0 _________________________________________________________________ dense_1 (Dense) (None, 10) 30730 ================================================================= Total params: 20,083,018 Trainable params: 20,083,018 Non-trainable params: 0 _________________________________________________________________
model.compile(
loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
model.fit(X_train, Y_train_oh, epochs=10)
Train on 100000 samples Epoch 1/10 100000/100000 [==============================] - 35s 347us/sample - loss: 1.2441 - accuracy: 0.5492 Epoch 2/10 100000/100000 [==============================] - 32s 324us/sample - loss: 0.8670 - accuracy: 0.6960 Epoch 3/10 100000/100000 [==============================] - 33s 326us/sample - loss: 0.7521 - accuracy: 0.7384 Epoch 4/10 100000/100000 [==============================] - 34s 337us/sample - loss: 0.6878 - accuracy: 0.7599 Epoch 5/10 100000/100000 [==============================] - 34s 335us/sample - loss: 0.6433 - accuracy: 0.7763 Epoch 6/10 100000/100000 [==============================] - 33s 332us/sample - loss: 0.6070 - accuracy: 0.7908 Epoch 7/10 100000/100000 [==============================] - 33s 332us/sample - loss: 0.5746 - accuracy: 0.8026 Epoch 8/10 100000/100000 [==============================] - 33s 329us/sample - loss: 0.5522 - accuracy: 0.8093 Epoch 9/10 100000/100000 [==============================] - 34s 342us/sample - loss: 0.5350 - accuracy: 0.8168 Epoch 10/10 100000/100000 [==============================] - 34s 336us/sample - loss: 0.5236 - accuracy: 0.8196
<tensorflow.python.keras.callbacks.History at 0x22577a74fc8>
fig, AX = plt.subplots(4, 8, sharex=True, sharey=True)
AX = [b for a in AX for b in a]
pred = model.predict(X_test)
np.random.seed(1234)
for ax in AX:
index = np.random.randint(Y_test.size)
# Prediction
A_ = pred[index]
Y_ = np.argmax(A_)
# Select sample image
imshow = X_test[index]
# Green square: predited correctly
# Red square: predicted wrongly
if Y_ == Y_test[index]:
imshow[-4:, -4:] = (0, 1, 0)
else:
imshow[-4:, -4:] = (1, 0, 0)
ax.imshow(imshow)
ax.set_title(f'{objects[Y_]} ({Y_})' + r' ; $A_i$ = {:.02f}'.format(float(A_[Y_])), fontsize=12)
print('Validation')
pred1 = model.predict(X_test)
print('Test:', np.sum((Y_test_oh - pred1)**2, axis=1).mean())
pred2 = model.predict(X_train)
print('Train:', np.sum((Y_train_oh - pred2)**2, axis=1).mean())
Validation Test: 0.2775033290645693 Train: 0.14203523197436863