I have data where each training example is a mix between sequential and non sequential.
For instance, my dataset currently returns something like:
inputs = (scalar_a, sequence_a, sequence_b) targets = sequence_c return inputs, targets
All sequences (a, b, c) are of same length but of course different examples in a batch may have different lengths.
My desired output is when I iterate through the dataloader I should get a tuple with the same structure except everything should be batched.
How do I go about doing this?