Finding the right shape for data initialization without padding a lot or without reducing data

[mint trail]

why are you thinking that the batch size needs to be 2^n? it seems like there is some confusion here

since you're using an rnn, I'm assume you're just doing this for learning and don't have many restrictions.

I would cast the problem as predicting M steps into the future given N steps from the past.

so if you have some data of shape data : [542_375], I would split it up as train_data = data[:400_000] (test_data gets the rest of the data). next, sample some index from the range (0, 400_000 - M - N -1). x_sample = data_train[index : index + N] and y_sample = data_train[index + N: index + N + M]

then if you have a batch size of B, you can just do this B times and stack them

I'm new here, but it probably isn't a good idea to post questions multiple times btw

[sweet notch]

thanks for the help a lot!🥹