mirror of
https://github.com/wassname/ray.git
synced 2026-07-12 11:59:20 +08:00
fix policy gradients for mujoco domains (#589)
This commit is contained in:
committed by
Robert Nishihara
parent
c5bc76193f
commit
679910496e
@@ -42,7 +42,7 @@ class Categorical(object):
|
||||
class DiagGaussian(object):
|
||||
def __init__(self, flat):
|
||||
self.flat = flat
|
||||
mean, logstd = tf.split(1, 2, flat)
|
||||
mean, logstd = tf.split(flat, 2, axis=1)
|
||||
self.mean = mean
|
||||
self.logstd = logstd
|
||||
self.std = tf.exp(logstd)
|
||||
|
||||
Reference in New Issue
Block a user