Testing out networks with ResNet reveals that we need to add an arg parse for encoder. Currently in trainNet.py the call to make_env doesn't accept an encoder at all.
Also it would make setup faster if certain combinations of arguments should be passed in (modularly) together:
--env='MiniWorld-LRoom-v0' \ --envPackage='miniworld_vae' \ --agent='MiniworldRandomAgent' \ --actenc='ContSpeedOnehotHDMiniworld'
(e.g. when needing to switch between minigrid and miniworld you'd need to change all four of these individually, but it would be nice to have presets for these...
... enter: hydra!
Testing out networks with ResNet reveals that we need to add an arg parse for encoder. Currently in
trainNet.pythe call tomake_envdoesn't accept an encoder at all.Also it would make setup faster if certain combinations of arguments should be passed in (modularly) together:
--env='MiniWorld-LRoom-v0' \ --envPackage='miniworld_vae' \ --agent='MiniworldRandomAgent' \ --actenc='ContSpeedOnehotHDMiniworld'(e.g. when needing to switch between
minigridandminiworldyou'd need to change all four of these individually, but it would be nice to have presets for these...... enter: hydra!