wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-09-09 11:32:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

Policy gradient example: record stats for tensorboard (#577)

Browse Source 
* add tf metrics

* comments

* fix network scopes

* add doc

* use format string

* fix trace level

* plot intermediate and final sgd stats

* add back a global step
This commit is contained in:
Eric Liang
2017-05-21 14:51:24 -07:00
committed by Philipp Moritz
parent c440010cbd
commit 06241daf61
7 changed files with 113 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/policy_gradient/reinforce/agent.py
+12 -9
View File
@@ -21,15 +21,18 @@ class Agent(object):
if preprocessor.shape is None:
preprocessor.shape = self.env.observation_space.shape
self.sess = tf.Session()
self.ppo = ProximalPolicyLoss(self.env.observation_space,
self.env.action_space, preprocessor, config,
self.sess)
self.optimizer = tf.train.AdamOptimizer(config["sgd_stepsize"])
self.train_op = self.optimizer.minimize(self.ppo.loss)
self.variables = ray.experimental.TensorFlowVariables(self.ppo.loss,
self.sess)
self.observation_filter = MeanStdFilter(preprocessor.shape, clip=None)
self.reward_filter = MeanStdFilter((), clip=5.0)
with tf.name_scope("policy_gradient/train"):
with tf.name_scope("proximal_policy_loss"):
self.ppo = ProximalPolicyLoss(self.env.observation_space,
self.env.action_space, preprocessor,
config, self.sess)
with tf.name_scope("adam_optimizer"):
self.optimizer = tf.train.AdamOptimizer(config["sgd_stepsize"])
self.train_op = self.optimizer.minimize(self.ppo.loss)
self.variables = ray.experimental.TensorFlowVariables(self.ppo.loss,
self.sess)
self.observation_filter = MeanStdFilter(preprocessor.shape, clip=None)
self.reward_filter = MeanStdFilter((), clip=5.0)
self.sess.run(tf.global_variables_initializer())
def get_weights(self):