Feature Extractor

This package has the ability to use custom feature extractors for the agent networks. When using a custom feature extractor the observations are first fed into the feature extractor and the result is then fed into the agent networks. Please see the stable-baselines3 section on feature extractors for more information on how this works.

This library is now also able to use more complex observation spaces, like image based observation spaces and observation spaces which are made up of scalar observations (normal observations) and image observations.

Both functionalities come from the feature extractors. Please see the class definitions of releso.feature_extractor.FeatureExtractor and releso.feature_extractor.CombinedExtractor for the possible parameters.

The image based observations are explained in more detail on the page Image based Observations.