Play

Play is the mode of using a baseline or checkpoint to load a policy and use it against an environment. There is no training involved. This mode is basically for enjoyment or evaluation purposes.

Run a Play Session

By default, ReleZoo runs in training mode. To change this behavior provide the option context.mode=play. However, play mode makes sense when using a checkpoint or a baseline since the idea is to watch a smart agent play!. So the full command shouls look like this:

relezoo-run context.mode=play context.checkpoints=baselines/reinforce/classic_control/cartpole/

Notes - Play mode only uses the environments@env_test property, so you only need to change this when playing against other environments. environments@env_train is ignored. - Play also generate logs for later inspection.

This will trigger the default environment and algorithm in play or inference mode using the provided checkpoint as loaded model. The rest of the options can be modified as well.

You must carefully select the correct combination of algorithms, environments and checkpoints, otherwise the results are not guaranteed.