width_shift_range=0.1,
height_shift_range=0.1,
zoom_range=0.1,
horizontal_flip=True)
datagen.fit(train_x)
#valid_x, valid_y = load_data('valid', 'train')
#model = load_model('model')
#model.save_weights('weight')
model = build_model()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
#model.load_weights('best_w')
history = History()
model.fit_generator(
datagen.flow(train_x, train_y, batch_size=128),
steps_per_epoch=len(train_x) / 128 * 8,
#steps_per_epoch = 1,
epochs=50,
#validation_data = (valid_x, valid_y),
callbacks=[history])
history.save('history')
model.save(model_path)
#model.save_weights('weight')
#print('\nvalid:', model.evaluate(valid_x, valid_y)[1])
exit()
history1 = History()
model = build_model()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit_generator(
datagen.flow(data1_x, data1_y, batch_size=128),
steps_per_epoch=len(data1_x) / 128 * 4,
#steps_per_epoch = 1,
epochs=50,
validation_data=(valid_x, valid_y),
callbacks=[history1])
#model.fit(data1_x, data1_y, batch_size = 128, epochs = 100, validation_data = (valid_x, valid_y), callbacks = [history1])
model.save('self_model')
history1.save('history1')
datagen.fit(data2_x)
history2 = History()
data2_y = to_categorical(model.predict_classes(data2_x, batch_size=128), 7)
model2 = build_model()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit_generator(
datagen.flow(data2_x, data2_y, batch_size=128),
steps_per_epoch=len(data2_x) / 128 * 4,
#steps_per_epoch = 1,
epochs=50,
rotation_range = 3,
width_shift_range = 0.1,
height_shift_range = 0.1,
zoom_range = 0.1,
horizontal_flip = True)
datagen.fit(train_x)
valid_x, valid_y = load_data('valid', 'train')
#model = load_model('model')
#model.save_weights('weight')
model = build_dnn_model()
model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])
#model.load_weights('best_w')
history = History()
model.fit_generator(
datagen.flow(train_x, train_y, batch_size = 128),
steps_per_epoch = len(train_x) / 128 * 2,
#steps_per_epoch = 1,
epochs = 50,
validation_data = (valid_x, valid_y),
callbacks = [history])
history.save('dnn_history')
#model.save(model_path)
#model.save_weights('weight')
print('\nvalid:', model.evaluate(valid_x, valid_y)[1])
exit()