czthehusky / competition_olympics-running

A guide to train your husky-like agents

AI3617 GTMAL Final Project

https://www.bilibili.com/video/BV1TY4y1x7u1

https://www.bilibili.com/video/BV1rF411G72L

Training

Current multithread settings are tested with two A100 GPUs with 40GB of memory each.

cd rl_trainer_rnn_mt

python main.py --mix_maps --reward_shaping

Resume training from a check point

cd rl_trainer_rnn_mt

python main.py --mix_maps --reward_shaping --check_point --actor_path '' --critic_path '' --ep ep_num --tc train_count

Self-play

Don't forget to prepare 17 opponents in './self_play_opponents' first. Feel free to choose them from your stored models.

cd rl_trainer_rnn_mt

python self_play.py --actor_path '' --critic_path '' --reward_shaping

Evaluation

cd evaluation

python eval.py