Closed bihanbihan closed 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]
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
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
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.