minst 관련 연습 소프트웨어

 from keras.datasets import mnist

from keras.utils import np_utils

from keras.models import Sequential

from keras.layers import Dense

 

(X_train, Y_train), (X_validation, Y_validation) = mnist.load_data()

 

X_train = X_train.reshape(X_train.shape[0], 784).astype('float64') / 255

X_validation = X_validation.reshape(X_validation.shape[0], 784).astype('float64') / 255

 

Y_train = np_utils.to_categorical(Y_train, 10)

Y_validation = np_utils.to_categorical(Y_validation, 10)

 

model = Sequential()

model.add(Dense(512, input_dim=784, activation='relu'))

model.add(Dense(10, activation='softmax'))

 

model.compile(loss='categorical_crossentropy', optimizer='adam',

              metrics=['accuracy'])

 

hist = model.fit(X_train, Y_train, validation_data=(X_validation, Y_validation),

          epochs=30, batch_size=500)

 

print('\nAccuracy: {:.4f}'.format(model.evaluate(X_validation, Y_validation)[1]))

 

import matplotlib.pyplot as plt

 

fig, loss_ax = plt.subplots()

acc_ax = loss_ax.twinx()

 

loss_ax.plot(hist.history['loss'], 'y', label='train loss')

loss_ax.plot(hist.history['val_loss'], 'r', label='val loss')

loss_ax.set_xlabel('epoch')

loss_ax.set_ylabel('loss')

loss_ax.legend(loc='upper left')

 

acc_ax.plot(hist.history['acc'], 'b', label='train acc')

acc_ax.plot(hist.history['val_acc'], 'g', label='val acc')

acc_ax.set_ylabel('accuracy')

acc_ax.legend(loc='upper left')

 

plt.show()