hiro/ray
1
0
Fork 0
mirror of https://github.com/vale981/ray synced 2025-03-06 02:21:39 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

[RLlib; docs] Clarify how MultiDiscrete spaces are encoded by default. (#23777)

Browse source
This commit is contained in:
Michael (Mike) Gelbart 2022-04-07 23:39:09 -07:00 • committed by GitHub
parent 7a46f5176a
commit 774b62b3c0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
doc/source/rllib/rllib-models.rst
View file

@ -29,8 +29,11 @@ observation space. Thereby, the following simple rules apply:
- Discrete observations are one-hot encoded, e.g. ``Discrete(3) and value=1 -> [0, 1, 0]``.
- MultiDiscrete observations are "multi" one-hot encoded,
e.g. ``MultiDiscrete([3, 4]) and value=[1, 0] -> [0 1 0 1 0 0 0]``.
- MultiDiscrete observations are encoded by one-hot encoding each discrete element
and then concatenating the respective one-hot encoded vectors.
e.g. ``MultiDiscrete([3, 4]) and value=[1, 3] -> [0 1 0 0 0 0 1]`` because
the first ``1`` is encoded as ``[0 1 0]`` and the second ``3`` is encoded as
``[0 0 0 1]``; these two vectors are then concatenated to ``[0 1 0 0 0 0 1]``.
- Tuple and Dict observations are flattened, thereby, Discrete and MultiDiscrete
sub-spaces are handled as described above.