1.0 KiB
1.0 KiB
tianshou
Tianshou(天授) is a reinforcement learning platform.
agent
Examples
Self-play Framework
core
Model
DQN, Policy-Value Network of AlphaGo Zero, PPO-specific, TROP-specific
Algorithm
Loss design
Actor-Critic (Variations), DQN (Variations), DDPG, TRPO, PPO
Optimization method
SGD, ADAM, TRPO, natural gradient, etc.
Planning
MCTS
data
Training style - Monte Carlo or Temporal Difference
Reward Reshaping/ Advantage Estimation Function
Importance weight
Multithread Read/Write
environment
DQN repeat frames etc.
simulator
Go, Othello/Reversi, Warzone
TODO
Search based method parallel.