Learning Homophilic Incentives in Sequential Social Dilemmas

Our method is built on PyMARL and Sequential Social Dilemma Games (SSDG). We use the implementation of Cleanup and Harvest Game from the SSDG repo. The implementation of our algorithm is based on independent Q-learning from the PyMARL repo.

Installation instructions

Set up a virtual environment and install the necessary packages using requirements.txt file.

conda create -n SSD python=3.7
conda activate SSD
pip install -r requirements.txt

Run an experiment

python3 src/main.py 
--config=homophily 
--env-config=cleanup 
with 
env_args.map=default5
env_args.num_agents=5
use_tensorboard=True 

For Cleanup with 10 agents, set env_args.map=default10 and env_args.num_agents=10. For Cleanup with 3 agents, set env_args.map=default3 and env_args.num_agents=3. For Harvest, set --env-config=harvest.

The config files contain default settings 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.

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 False by default. The frequency of saving models can be adjusted using save_model_interval. Models will be saved in the result directory, under the folder called models. The directory corresponding to 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 replays

save_replay option allows saving replays of models which are loaded from checkpoint_path. Once the model is successfully loaded, test_nepisode number of episodes are run in the test mode.

python3 src/main.py 
--config=homophily 
--env-config=cleanup 
with 
env_args.map=default5
env_args.num_agents=5
use_tensorboard=False 
save_replay=True 
env_args.is_replay=True 
test_nepisode=1 
checkpoint_path="results/models/model_file_name"

The replays can be found in results/replays/ folder.