Multiagent ASN

Installation instructions

This project based on SMAC and pymarl

OS: Ubuntu 14 or 16

Install StarCraftII:

chmod 755 ./install_sc2.sh
./install_sc2.sh
tar -xvf SMAC_Maps_supply.tar.gz
mv SMAC_Maps/* 3rdparty/StarCraftII/Maps/SMAC_Maps/

You must install anaconda before installing code env.

Install Code Env:

conda create -n ASN python=3.6
source activate ASN
conda install certifi==2018.8.24
pip install -r requirements.txt

Run an experiment

Activate Code Env:

source activate ASN
export PYTHONPATH=. 

Then:

8m:

IQL

Vanilla

python src/main.py --config=iql_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Attention

python src/main.py --config=iql_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_attention' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Entity-Attention

python src/main.py --config=iql_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_entity_attention' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Dueling

python src/main.py --config=iql_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_dueling' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

ASN

python src/main.py --config=iql_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='asn_rnn' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

VDN

Vanilla

python src/main.py --config=vdn_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Attention

python src/main.py --config=vdn_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_attention' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Entity-Attention

python src/main.py --config=vdn_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_entity_attention' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

Dueling

python src/main.py --config=vdn_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='dense_rnn_dueling' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

ASN

python src/main.py --config=vdn_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 agent='asn_rnn' legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100

QMIX

Vanilla

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 env='sc2' seed=100 agent='dense_rnn'

Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 env='sc2' seed=100 agent='dense_rnn_attention'

Entity-Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 env='sc2' seed=100 agent='dense_rnn_entity_attention'

Dueling

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 env='sc2' seed=100 agent='dense_rnn_dueling'

ASN

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 env='sc2' seed=100 agent='asn_rnn'

-1-Paddings

QMIX

Vanilla

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn' env='sc2_not_0'

Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_attention' env='sc2_not_0'

Entity-Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_entity_attention' env='sc2_not_0'

Dueling

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_dueling' env='sc2_not_0'

ASN

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='asn_rnn' env='sc2_not_0'

1-Paddings

QMIX

Vanilla

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn' env='sc2_set_1'

Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_attention' env='sc2_set_1'

Entity-Attention

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_entity_attention' env='sc2_set_1'

Dueling

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='dense_rnn_dueling' env='sc2_set_1'

ASN

python src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=8m_8m agents_num=8 enemies_num=8 legal_action=False batch_size_run=1 use_tensorboard=True save_model=True runner_log_interval=2000 seed=100 agent='asn_rnn' env='sc2_set_1'