NDQ: Learning Nearly Decomposable Value Functions with Communication Minimization

Note

This codebase accompanies paper Learning Nearly Decomposable Value Functions with Communication Minimization, and is based on PyMARL and SMAC codebases which are open-sourced.

The implementation of the following methods can also be found in this codebase, which are finished by the authors of PyMARL:

• QMIX: QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning
• COMA: Counterfactual Multi-Agent Policy Gradients
• VDN: Value-Decomposition Networks For Cooperative Multi-Agent Learning
• IQL: Independent Q-Learning

Build the Dockerfile using

cd docker
bash build.sh

Set up StarCraft II and SMAC:

bash install_sc2.sh

This will download SC2 into the 3rdparty folder and copy the maps necessary to run over.

The requirements.txt file can be used to install the necessary packages into a virtual environment (not recomended).

Run an experiment

The following command train NDQ on the didactic task hallway.

python3 src/main.py 
--config=categorical_qmix
--env-config=join1
with
env_args.n_agents=2
env_args.state_numbers=[6,6]
obs_last_action=False
comm_embed_dim=3
c_beta=0.1
comm_beta=1e-2
comm_entropy_beta=0.
batch_size_run=16
t_max=2e7
local_results_path=$DATA_PATH
is_cur_mu=True
is_rank_cut_mu=True
runner="parallel_x"
test_interval=100000

The config files act as defaults for an algorithm or environment.

They are all located in src/config. --config refers to the config files in src/config/algs --env-config refers to the config files in src/config/envs

To train NDQ on SC2 tasks, run the following command:

--config=categorical_qmix
--env-config=sc2
with
env_args.map_name=bane_vs_hM
env_args.sight_range=2
env_args.shoot_range=2
env_args.obs_all_health=False
env_args.obs_enemy_health=False
comm_embed_dim=3
c_beta=0.1
comm_beta=0.0001
comm_entropy_beta=0.0
batch_size_run=16
runner="parallel_x"

SMAC maps can be found in src/smac_plus/sc2_maps/.

All results will be stored in the Results folder.

Saving and loading learnt models

Saving models

You can save the learnt models to disk by setting save_model = True, which is set to False by default. The frequency of saving models can be adjusted using save_model_interval configuration. Models will be saved in the result directory, under the folder called models. The directory corresponding each run will contain models saved throughout the experiment, each within a folder corresponding to the number of timesteps passed since starting the learning process.

Loading models

Learnt models can be loaded using the checkpoint_path parameter, after which the learning will proceed from the corresponding timestep.

Watching StarCraft II replays

save_replay option allows saving replays of models which are loaded using checkpoint_path. Once the model is successfully loaded, test_nepisode number of episodes are run on the test mode and a .SC2Replay file is saved in the Replay directory of StarCraft II. Please make sure to use the episode runner if you wish to save a replay, i.e., runner=episode. The name of the saved replay file starts with the given env_args.save_replay_prefix (map_name if empty), followed by the current timestamp.

The saved replays can be watched by double-clicking on them or using the following command:

python -m pysc2.bin.play --norender --rgb_minimap_size 0 --replay NAME.SC2Replay

Note: Replays cannot be watched using the Linux version of StarCraft II. Please use either the Mac or Windows version of the StarCraft II client.