learning_from_play
learning_from_play copied to clipboard
Dataloader optimisation
- mapping transform BEFORE windowing
- var seq lens and goal tiling are computationally expensive (we ought to do on device) so I've removed from dataloader
- multi-threaded tfrecords mapper (NOT the IO, this has to be single threaded)
- action validation also taxing, it could be risky removing but it's also too slow. We should validate our data once off beforehand