class TraditionalRLRewardPredictor():
"""Predictor that always returns the true reward provided by the environment."""
def __init__(self, summary_writer):
self.agent_logger = AgentLogger(summary_writer)
def predict_reward(self, path):
self.agent_logger.log_episode(path)
return path["original_rewards"]
def path_callback(self, path):
pass
class TraditionalRLRewardPredictor(object):
"""Predictor that always returns the true reward provided by the environment."""
def __init__(self, summary_writer):
self.agent_logger = AgentLogger(summary_writer)
def predict_reward(self, path):
self.agent_logger.log_episode(
path) # <-- This may cause problems in future versions of Teacher.
return path["original_rewards"]
def path_callback(self, path):
pass
class TraditionalRLRewardPredictor(object):
"""Predictor that always returns the true reward provided by the environment."""
def __init__(self, summary_writer, agent_logger=None):
self.agent_logger = AgentLogger(
summary_writer) if agent_logger is None else agent_logger
self.sess = None
self.comparison_collector = None
def predict_reward(self, path):
self.agent_logger.log_episode(
path) # <-- This may cause problems in future versions of Teacher.
return path["original_rewards"]
def path_callback(self, path):
pass
def save_session(self, path, global_step):
self.agent_logger.save(path, global_step)
with open(os.path.join(path, 'iteration.pkl'), 'wb') as f:
pickle.dump(global_step, f)
def load_session(self, path):
pass