I think a more natural interface for the chunking capability would be to just set the chunk_size parameter. If we iterate through the whole dataset, then we can determine the number of steps based on the chunk_size (clipping off any extra samples that aren't in an integer multiple of chunk_size). Some logic will also be needed to handle the case of mismatched sizes between x and y. I think the natural choice is to iterate over the larger dataset chunks and cycle on the smaller one's chunks.
All the documentation should be updated to reflect the new design.
I think a more natural interface for the chunking capability would be to just set the chunk_size parameter. If we iterate through the whole dataset, then we can determine the number of steps based on the chunk_size (clipping off any extra samples that aren't in an integer multiple of chunk_size). Some logic will also be needed to handle the case of mismatched sizes between x and y. I think the natural choice is to iterate over the larger dataset chunks and cycle on the smaller one's chunks.
All the documentation should be updated to reflect the new design.