mila-iqia/babyai

default hyperparameters don't make sense for both IL and RL

rizar opened this issue · 2 comments

rizar commented

For example, --save-interval=1 makes sense for IL but not for RL. --batch-size=1280 is the one we used for RL, not for IL.

We should somehow allow these scripts to have different hyperparameters by default.

The --save-interval is now different. Dima, do you have a suggestion for an appropriate batch size for IL?

rizar commented

For "small" model I used --batch-size=256, for "large" model I used --batch-size=128. We could switch to using the same batch size for both settings now, since after the bot was optimized, the demonstrations have become shorter.