3_CNN_101.py

import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np

# Import data

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

x_train = x_train[..., tf.newaxis].astype('float32')
x_test = x_test[..., tf.newaxis].astype('float32')

train_ds = tf.data.Dataset.from_tensor_slices((x_train, y_train)).shuffle(10000).batch(100)
test_ds = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(100)

# Define Multilayer Perceptron model defined with the tf.keras module

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(32, kernel_size=(3, 3), strides=(1,1), padding='same', input_shape=(28, 28, 1)))
model.add(tf.keras.layers.Conv2D(64, kernel_size=(3, 3), activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=(2, 2)))
model.add(tf.keras.layers.Dropout(0.25))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(10, activation='softmax'))

model.summary()

# Run model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
history = model.fit(x_train, y_train, batch_size=100, epochs=2)


print(history.history)