saymrwulf/stable-baselines3
1
0
Fork 0
mirror of https://github.com/saymrwulf/stable-baselines3.git synced 2026-06-05 00:00:04 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add a note for squashed gaussian

Browse source
This commit is contained in:
Antonin RAFFIN 2019-09-22 14:15:22 +02:00
parent b157b4465a
commit 1bd2e42995
1 changed files with 2 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
torchy_baselines/common/distributions.py
View file

@ -103,6 +103,8 @@ class SquashedDiagGaussianDistribution(DiagGaussianDistribution):
# Inverse tanh
# Naive implementation (not stable): 0.5 * torch.log((1 + x ) / (1 - x))
# We use numpy to avoid numerical instability
# TODO: store the gaussian action because of the action added
# this would avoid also inverting the tanh
if gaussian_action is None:
gaussian_action = th.from_numpy(np.arctanh(action.cpu().numpy())).to(action.device)