Update the action space of PointMaze to be [-1, 1]^2 #101
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Related to #37
All envs have an action space being a cartesian product of [-1, 1], except PointMaze that has [-0.1, 0.1]. This PR fixes this by moving the action space of PointMaze to [-1, 1]^2.
This has a small impact on the optimization process. Illustrated by the following plots.
Optim with MAP-Elites and the previous action space:
Optim with MAP-Elites and the new action space:
The slight difference in the optimization process arises because the previous implementation was feeding arbitrarily large actions (i.e. in the range of -1 and 1) where the action was then clipped to defined min max of the environment (i.e. -0.1 and 0.1). The new implementation standardizes the action space to force min max of the environment to be -1, 1 and then does the scaling to -0.1 and 0.1 internally within the environment. We can expect more similar results to the new implementation if we rescaled the outputs of the policy to -0.1 and 0.1 before passing it into the old environment implementation.