Keras shuffling of multiple inputs

问题:

I have a question regarding the shuffling of multiple inputs during the training phase of a model. Given the following call:

model.fit([input_a_0, input_a_1, input_b_0, input_b_1], y_train,
          batch_size=8,
          epochs=4,
          shuffle=True)

All the inputs need to be shuffled in the exact same order so that they align correctly. Now my question is how Keras handles this? Are they all shuffled in the same way (using the same permutation index) or is each input shuffled independently of the other inputs?

回答1:

Naturally Keras will shuffle them correctly, otherwise there would be a huge number of complaints and open issues about how Keras couldn't reach any good results.