I try to rerun experiment by using this repo like below:
python3 src/main.py --no-mongo --config=ppo_central_V --env-config=sc2_gen_terran with env_args.capability_config.n_units=20 env_args.capability_config.n_enemies=20 use_wandb=False mini_epochs_actor=10 clip_range=0.1 detach_every=20
And I found test_battle_won_mean keep zero when training step reach 2M.
I try to rerun experiment by using this repo like below: python3 src/main.py --no-mongo --config=ppo_central_V --env-config=sc2_gen_terran with env_args.capability_config.n_units=20 env_args.capability_config.n_enemies=20 use_wandb=False mini_epochs_actor=10 clip_range=0.1 detach_every=20
And I found test_battle_won_mean keep zero when training step reach 2M.