Link: StarCraft 2 Reinforcement Learning
Random agent
: as the name shows, it does things randomlyA3C
: A3C method implemented in our repository branchA2C
: A2C method implemented in our repository branch
Metrics | Random agent | A3C | A2C |
---|---|---|---|
Mean Score | 1.81 | 3.4 | 13.69 |
Max Score | 46 | 71 | 81 |
See this repo
simon-a2c
: A2C method referenced from a fork of simonmeister's repo, abbreviation assimon-a2c
xhujoy-a3c
: a fork of Xiaowei Hu's reporandom agent
: as the name shows, it does things randomly
Map | Training iters | simon-a2c | random agent | DeepMind human |
---|---|---|---|---|
(mean) DefeatRoaches | 1000×16 = 16k | 5.74 | -4 | 41 |
( max ) DefeatRoaches | Same | 61.00 | 11 | 81 |
(mean) DefeatZerglingsAndBanelings | 500×8 = 4k | 24.8 | 18.1 | 729 |
( max ) DefeatZerglingsAndBanelings | Same | 108.00 | 72 | 757 |
(mean) MoveToBeacon | 5000×6 = 30k | 25.26 | 1.2 | 26 |
( max ) MoveToBeacon | Same | 30.00 | 3 | 28 |
(mean) CollectMineralShards | 2000×4 = 8k | 23.84 | 18.5 | 133 |
( max ) CollectMineralShards | Same | 39.00 | 32 | 142 |
(mean) FindAndDefeatZerglings | 500×8 = 4k | 4.82 | 5 | 46 |
( max ) FindAndDefeatZerglings | Same | 18.00 | 15 | 49 |