Skip to content

[rllib] MultiDiscrete action distributions aren't handled correctly in all algorithms #4866

@ericl

Description

@ericl

Describe the problem

Currently only IMPALA and APPO handle MultiDiscrete properly. We should fix this or raise an error telling users to use Tuple instead.

As a workaround, you can use Tuple[Discrete, Discrete, ...] instead of MultiDiscrete.

Source code / logs

This is from https://groups.google.com/forum/#!topic/ray-dev/RaqWO89Rl6E

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions