I am asking this model fit( ) function.

fit(x=array_x, y=array_y, batch_size=32, epochs=10)

The question is straightforward:

Whether fit( ) will automatically shuffle the input dataset? (array_x, array_y in this case)

I ask this question because I find my fit( ) and fit_generator( ) has different outputs, with same input. My generator feeds inputs to fit_generator( ) in order. I guess the reason is about shuffle.


Yes, by default it does shuffle.

Here is the documentation.

The default call signature:

fit(x=None, y=None, batch_size=None, epochs=1, verbose=1, callbacks=None,
    validation_split=0.0, validation_data=None, shuffle=True, 
    lass_weight=None, sample_weight=None, initial_epoch=0,
    steps_per_epoch=None, validation_steps=None, validation_freq=1,
    ax_queue_size=10, workers=1, use_multiprocessing=False)

where the description for shuffle is:

shuffle: Boolean (whether to shuffle the training data before each epoch) or str (for 'batch'). This argument is ignored when x is a generator. 'batch' is a special option for dealing with the limitations of HDF5 data; it shuffles in batch-sized chunks. Has no effect when steps_per_epoch is not None.

This means it will not shuffle if you have set a value for steps_per_epoch.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.