I’m using 2 generator to generate input for each step and need to update input at certain training step, but prefetched data often get a bit messed up when changing input. I think of 2 way to solve problem:
- Remove data that has been called to input pipeline by prefetch (since I use generator and need to update input at certain training step)
- Stop prefetching when it prefetch enough data at training step k
Is there any straight-forward way solve 1 of 2 problems above?