如何保存Keras回归模型？

Question

如何在训练后保存模型重量？

Keras提供:

model.save( 'weights.h5')`

模型对象由build_fn属性函数初始化,如何进行保存？

def model():
    model = Sequential()
    model.add(Dense(10, activation='relu', input_dim=5))
    model.add(Dense(5, activation='relu'))
    model.add(Dense(1, kernel_initializer='normal'))
    model.compile(loss='mean_squared_error', optimizer='adam')
    return model


if __name__ == '__main__':
`

    X, Y = process_data()

    print('Dataset Samples: {}'.format(len(Y)))

    model = KerasRegressor(build_fn=model,
            epochs=10,
            batch_size=10,
            verbose=1)


    kfold = KFold(n_splits=2, random_state=seed)

    results = cross_val_score(model, X, Y, cv=kfold)

    print('Results: {0}.2f ({1}.2f MSE'.format(results.mean(), results.std()))

Answer 1

cross_val_score克隆提供的估算器,将它们放在训练折叠上,在测试折叠上得分.基本上,你的实际模型尚未安装.

首先,您需要在数据上拟合模型:

model.fit(X, Y)

然后,您可以使用基础model属性(实际存储keras模型)来调用save()or save_weights()方法.

model.model.save('saved_model.h5')

现在,当您想再次加载模型时,请执行以下操作:

from keras.models import load_model

# Instantiate the model as you please (we are not going to use this)
model2 = KerasRegressor(build_fn=model_build_fn, epochs=10, batch_size=10, verbose=1)

# This is where you load the actual saved model into new variable.
model2.model = load_model('hh.h5')

# Now you can use this to predict on new data (without fitting model2, because it uses the older saved model)
model2.predict(X)