gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 19:20:19 +01:00
Code Issues Projects Releases Wiki Activity

Release 0.9

Browse Source 
Main changes are detailed below:

New features -
* CARLA 0.7 simulator integration
* Human control of the game play
* Recording of human game play and storing / loading the replay buffer
* Behavioral cloning agent and presets
* Golden tests for several presets
* Selecting between deep / shallow image embedders
* Rendering through pygame (with some boost in performance)

API changes -
* Improved environment wrapper API
* Added an evaluate flag to allow convenient evaluation of existing checkpoints
* Improve frameskip definition in Gym

Bug fixes -
* Fixed loading of checkpoints for agents with more than one network
* Fixed the N Step Q learning agent python3 compatibility
This commit is contained in:
Itai Caspi
2017-12-19 19:27:16 +02:00
committed by GitHub
parent 11faf19649
commit 125c7ee38d
41 changed files with 1713 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
agents/n_step_q_agent.py
View File

@@ -45,7 +45,7 @@ class NStepQAgent(ValueOptimizationAgent, PolicyOptimizationAgent):
# 1-Step Q learning
q_st_plus_1 = self.main_network.target_network.predict(next_states)
for i in reversed(xrange(num_transitions)):
for i in reversed(range(num_transitions)):
state_value_head_targets[i][actions[i]] = \
rewards[i] + (1.0 - game_overs[i]) * self.tp.agent.discount * np.max(q_st_plus_1[i], 0)
@@ -56,7 +56,7 @@ class NStepQAgent(ValueOptimizationAgent, PolicyOptimizationAgent):
else:
R = np.max(self.main_network.target_network.predict(np.expand_dims(next_states[-1], 0)))
for i in reversed(xrange(num_transitions)):
for i in reversed(range(num_transitions)):
R = rewards[i] + self.tp.agent.discount * R
state_value_head_targets[i][actions[i]] = R