search

opendilab / DI-engine

OpenDILab Decision AI Engine. The Most Comprehensive Reinforcement Learning Framework B.P.
https://di-engine-docs.readthedocs.io
Apache License 2.0
3k stars 366 forks source link

question for SMAC #792

Closed bihanbihan closed 5 months ago

bihanbihan commented 5 months ago

when i train smac_5m6m_mappo_config.py,smac_5m6m_masac_config.py,and smac_5m6m_madqn_config.py.how can i get the battle win rate of these methods?i dont get the battle win rate when i train these models,and in eval period,it also only have the boolean result for battle win or battle false.

PaParaZz1 commented 5 months ago

The win rate is calculated by the average result of n_evaluator_episode games' boolean results (win/lose).

For example, if n_evaluator_episode=4, the win rate could be 0.75 when games' boolean results are [1, 0, 1, 1]

bihanbihan commented 5 months ago

eval_episode_return [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0,
0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0,
1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0,
1.0, 1.0] this is the part of training log,so is it mean the win rate is calculate by eval_episode_return? 17/32=0.53125

PaParaZz1 commented 5 months ago

eval_episode_return [0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0] this is the part of training log,so is it mean the win rate is calculate by eval_episode_return? 17/32=0.53125

Yes, the win rate is calculated by this rule and shown as eval_episode_return