lesson 11. Dropout.py

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.datasets import mnist         # библиотека базы выборок Mnist
from tensorflow import keras
from tensorflow.keras.layers import Dense, Flatten, Dropout

(x_train, y_train), (x_test, y_test) = mnist.load_data()

# стандартизация входных данных
x_train = x_train / 255
x_test = x_test / 255

y_train_cat = keras.utils.to_categorical(y_train, 10)
y_test_cat = keras.utils.to_categorical(y_test, 10)

limit = 5000
x_train_data = x_train[:limit]
y_train_data = y_train_cat[:limit]

x_valid = x_train[limit:limit*2]
y_valid = y_train_cat[limit:limit*2]

model = keras.Sequential([
    Flatten(input_shape=(28, 28, 1)),
    Dense(300, activation='relu'),
    Dropout(0.8),
    Dense(10, activation='softmax') ])

# print(model.summary())      # вывод структуры НС в консоль

model.compile(optimizer='adam',
             loss='categorical_crossentropy',
             metrics=['accuracy'])


his = model.fit(x_train_data, y_train_data, epochs=50, batch_size=32, validation_data=(x_valid, y_valid))


plt.plot(his.history['loss'])
plt.plot(his.history['val_loss'])
plt.show()