gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 03:30:19 +01:00
Code Issues Projects Releases Wiki Activity

bug fixes for OPE (#311)

Browse Source
This commit is contained in:
Gal Leibovich
2019-05-21 16:39:11 +03:00
committed by GitHub
parent 85d70dd7d5
commit acceb03ac0
8 changed files with 38 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
rl_coach/agents/agent.py
View File

@@ -697,7 +697,7 @@ class Agent(AgentInterface):
# we either go sequentially through the entire replay buffer in the batch RL mode,
# or sample randomly for the basic RL case.
training_schedule = self.call_memory('get_shuffled_data_generator', batch_size) if \
training_schedule = self.call_memory('get_shuffled_training_data_generator', batch_size) if \
self.ap.is_batch_rl_training else [self.call_memory('sample', batch_size) for _ in
range(self.ap.algorithm.num_consecutive_training_steps)]

2
rl_coach/agents/ddqn_bcq_agent.py
View File

@@ -155,7 +155,7 @@ class DDQNBCQAgent(DQNAgent):
reward_model_loss = 0
imitation_model_loss = 0
total_transitions_processed = 0
for i, batch in enumerate(self.call_memory('get_shuffled_data_generator', batch_size)):
for i, batch in enumerate(self.call_memory('get_shuffled_training_data_generator', batch_size)):
batch = Batch(batch)
# reward model

2
rl_coach/agents/value_optimization_agent.py
View File

@@ -164,7 +164,7 @@ class ValueOptimizationAgent(Agent):
for epoch in range(epochs):
loss = 0
total_transitions_processed = 0
for i, batch in enumerate(self.call_memory('get_shuffled_data_generator', batch_size)):
for i, batch in enumerate(self.call_memory('get_shuffled_training_data_generator', batch_size)):
batch = Batch(batch)
loss += self.get_reward_model_loss(batch)
total_transitions_processed += batch.size