When gaussian exploration error is added to the action, it might step outside the [0, 1] and [2, 5] domains, we need to clamp that using torch.clamp or something similar.
It would be good if we can let the Agent class itself handle the exploration as well, instead of handling it in the training loop. We would add an option to enable or disable it.
When gaussian exploration error is added to the action, it might step outside the
[0, 1]and[2, 5]domains, we need to clamp that usingtorch.clampor something similar.It would be good if we can let the
Agentclass itself handle the exploration as well, instead of handling it in the training loop. We would add an option to enable or disable it.