1. 08 9月, 2021 1 次提交
    • S
      feature(nyz): add supervised learning image classification training demo (#27) · 11cc97e8
      Swain 提交于
      * feature(nyz): add resnet for cv sl task
      
      * feature(nyz): add imagenet classification dataset and adapt compile config for sl
      
      * feature(nyz): add naive image training entry demo
      
      * style(nyz): polish image cls train log
      
      * polish(nyz): polish multi gpu training setting
      
      * feature(nyz): add nn training bp and update async execution
      
      * feature(nyz): add distributed sampler for different dist backend
      
      * fix(nyz): fix compile config collector and buffer compatibility problem
      
      * style(nyz): correct yapf format
      
      * fix(nyz): fix env manager compile config compatibility bug
      
      * refactor(nyz): abstarct ISerialEvaluator and rename serial evaluation implementation
      
      * refactor(nyz): refactor collector name
      
      * feature(nyz): add metric evaluator and image cls acc metric eval demo
      
      * fix(nyz): fix cuda and multi gpu bug in image cls demo
      11cc97e8
  2. 02 9月, 2021 1 次提交
    • S
      League Metric (#22) · d24f1f3d
      Swain 提交于
      * feature(nyz): add trueskill as league metric, naive elo calculator, fix game_env info bug
      
      * fix(nyz): fix league player mutate bug
      
      * fix(nyz): fix league unittest bug
      
      * feature(nyz): add elo ranking in league metric env
      
      * polish(nyz): modify fixed eval policy and trueskill init
      
      * feature(nyz): add init main player in evaluation and fix stop_value bug
      
      * style(nyz): rename test_league_metric to avoid pyc cache bug
      d24f1f3d
  3. 03 8月, 2021 1 次提交
    • S
      serial training league demo (#12) · 73295c22
      Swain 提交于
      * feature(nyz): add naive 1v1 two player demo
      
      * feature(nyz): add 1v1 evaluator and 2 rule-based policy for evaluation
      
      * feature(nyz): modify game env and adjust hyper-param
      
      * feature(nyz): add naive league training multi player demo
      
      * feature(nyz): enable force snapshot to support init historical league player; finish league demo basic code
      
      * feature(nyz): modify selfplay demo and add two type game env
      
      * style(nyz): correct format style
      
      * polish(nyz): correct format style and adapt league demo main
      
      * feature(nyz): add league payoff viz and enable payoff update in league demo
      
      * feature(nyz): modify win rate calculation with draws
      
      * test(nyz): fix one vs one league test compatibility bug
      
      * test(nyz): add selfplay and league demo into unittest and algotest
      
      * style(nyz): correct format
      
      * hotfix(nyz): fix ppo continuous comatibility bug
      73295c22