Precompute log probability in PPO

chainerrl/initializers/__init__.py

@@ -1,2 +1,5 @@
 from chainerrl.initializers.constant import VarianceScalingConstant  # NOQA
-from chainerrl.initializers.normal import LeCunNormal  # NOQA
+
+# LeCunNormal was merged into Chainer v3, thus removed from ChainerRL.
+# For backward compatibility, it is still imported in this namespace.
+from chainer.initializers import LeCunNormal  # NOQA

chainerrl/links/mlp.py

@@ -8,10 +8,9 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 
-from chainerrl.initializers import LeCunNormal
-
 
 class MLP(chainer.Chain):
     """Multi-Layer Perceptron"""

chainerrl/links/mlp_bn.py

@@ -8,10 +8,9 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 
-from chainerrl.initializers import LeCunNormal
-
 
 class LinearBN(chainer.Chain):
     """Linear layer with BatchNormalization."""

chainerrl/policies/deterministic_policy.py

@@ -10,11 +10,11 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 
 from chainerrl import distribution
 from chainerrl.functions.bound_by_tanh import bound_by_tanh
-from chainerrl.initializers import LeCunNormal
 from chainerrl.links.mlp import MLP
 from chainerrl.links.mlp_bn import MLPBN
 from chainerrl.policy import Policy

chainerrl/policies/gaussian_policy.py

@@ -10,12 +10,12 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 import numpy as np
 
 from chainerrl import distribution
 from chainerrl.functions.bound_by_tanh import bound_by_tanh
-from chainerrl.initializers import LeCunNormal
 from chainerrl import links
 from chainerrl.policy import Policy
 

chainerrl/q_functions/state_action_q_functions.py

@@ -8,9 +8,9 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 
-from chainerrl.initializers import LeCunNormal
 from chainerrl.links.mlp import MLP
 from chainerrl.links.mlp_bn import MLPBN
 from chainerrl.q_function import StateActionQFunction

examples/ale/train_a3c_ale.py

@@ -100,8 +100,9 @@ def main():
     logging.basicConfig(level=args.logging_level)
 
     # Set a random seed used in ChainerRL.
-    # If you use more than one processes, the results will be no longer
-    # deterministic even with the same random seed.
+    # If you use more than one process (i.e. processes > 1),
+    # the results will be no longer be deterministic
+    # even with the same random seed.
     misc.set_random_seed(args.seed)
 
     # Set different random seeds for different subprocesses.

examples/gym/train_acer_gym.py

@@ -13,6 +13,7 @@
 
 import chainer
 from chainer import functions as F
+from chainer.initializers import LeCunNormal
 from chainer import links as L
 import gym
 from gym import spaces
@@ -24,7 +25,6 @@
 from chainerrl.agents import acer
 from chainerrl.distribution import SoftmaxDistribution
 from chainerrl import experiments
-from chainerrl.initializers import LeCunNormal
 from chainerrl import links
 from chainerrl import misc
 from chainerrl.optimizers import rmsprop_async