keras - Proper number of steps_per_epoch when shuffle samples in batches

Question

I am using a generator in my LSTM keras model. In the generator I would like to shuffle samples using shuffle = True. My question is that when I shuffle data, what should be the number of steps_per_epoch so that during one epoch the model could be trained based on the entire train set. Since samples in batches are shuffled, using the traditional number total_num_samples/batch_size would not allow us to make sure that we use the entire training data over one epoch. Thanks for your thoughts in advance.

question from:https://stackoverflow.com/questions/66049565/proper-number-of-steps-per-epoch-when-shuffle-samples-in-batches

Answer 1

Waitting for answers