使用 Keras 在 Tensorflow 2.0 中无需训练即可保存模型

Question

I use Tensorflow 2.0 and the Keras sequential API to build a model. I would like to save an untrained model for baseline comparison. How do I do that? I tried to set epochs=0 and period=0 which did not work.

Here is my code:

network = NeuralNetwork() # sequential neural network
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath='./models/', save_weights_only=False, period=0)
network.fit(x_train, y_train, epochs=0, callbacks=[cp_callback])

How can I save an untrained model?

Answer 1

You simply have to call model.save after instanciation of your model :

network = NeuralNetwork() # sequential neural network
network.save("/your/path/.h5") 

ModelCheckpoint can only be called as a callback, so during the training.