Repository files navigation

• These results are from only 32 threads.
• A total of 32 CPUs were used, 4 environments were configured for each game type, and a total of 8 games were learned.
• Tensorflow Implementation
• Use DQN model to inference action
• Use distributed tensorflow to implement Actor
• Training with 1 day
• Same parameter ofpaper

start learning rate = 0.0006end learning rate = 0learning frame = 1e6gradient clip norm = 40trajectory = 20batch size = 32reward clipping = -1 ~ 1

tensorflow==1.14.0gym[atari]numpytensorboardXopencv-python

• showstart.sh
• Learning 8 types of games at a time, one of which uses 4 environments.

Breakout	Pong	Seaquest	Space-Invader
Boxing	Star-Gunner	Kung-Fu	Demon

abs_one	soft_asymmetric

abs_one
soft_asymmetric

• Above Blocks are ignored.
• Ball and Bar are attentioned.
• Empty space are attentioned because of less trained.

• Only CPU Training method
• Distributed tensorflow
• Model fix for preventing collapsed
• Reward Clipping Experiment
• Parameter copying from global learner
• Add Relational Reinforcement Learning
• Add Action information to Model
• Multi Task Learning
• Add Recurrent Model
• Training on GPU, Inference on CPU

1. IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
2. deepmind/scalable_agent
3. Asynchronous_Advatnage_Actor_Critic