Keras笔记

瞎写的

Keras函数式API

构造张量

inputs = Input(shape=(64,))

创建层

dense = layers.Dense(32, activation='relu')

output_tensor = dense(input_tensor)

输出也是一个张量

回调函数等

EarlyStopping中断训练, ModelCheckpoint保存模型,如

callbacks_list = [
keras.callbacks.EarlyStopping(
monitor='acc',
patience=1,
),
keras.callbacks.ModelCheckpoint(
filepath='my_model.h5',
monitor='val_loss',
save_best_only=True,
)
]

改变学习率,ReduceLROnPlateau函数

callbacks_list = [
keras.callbacks.ReduceLROnPlateau(
monitor='val_loss'
factor=0.1,
patience=10,
)
]
model.fit(x, y,
epochs=10,
batch_size=32,
callbacks=callbacks_list,
validation_data=(x_val, y_val))

参考

1. keras官方文档
2. Python深度学习