transactive_control

Code meant to support and simulate the Social Game that will be launched in 2020. Elements of transactive control and behavioral engineering will be tested and designed here

Installation

Clone the repo
Install dvc (with google drive support)

On linux this is pip install 'dvc[gdrive]'

Install Docker, if you have not already.
Run python3 -m dvc remote add -d gdrive gdrive://1qaTn6IYd3cpiyJegDwwEhZ3LwrujK3_x
Run python3 -m dvc pull

Usage

Run ./run.sh from the root of the repo. This will put you in a shell in the docker container with the rl_algos/logs directory mounted
Run python3 StableBaselines.py sac test_experiment in the docker container to start an experiment with the name test_experiment
Run tensorboard --logdir rl_algos/logs from outside the docker container to view the logs

12/20/2020

This repository has been cleaned and updated for use. It contains: (1) The OpenAI gym environment "OfficeLearn", in the "gym-socialgame" folder, and (2) implementations of Reinforcement learning algorithms in "rl_algos" folder. In the "simulations" folder are various datasets for setting up and training models associated with the gym simulation.

9/1/2020

The most recent running of the code involves navigating to the rl_algos/ directory, then running the python command for the vanilla version:

python StableBaselines.py sac

Adding in the planning model can be done with the following flags:

python StableBaselines.py sac --planning_steps=10 --planning_model=Oracle --num_steps=10000

Please see transactive_control/gym-socialgame/gym_socialgame/envs for files pertaining to the setup of the environment. The socialgame_env.py contains a lot of the information necessary for understanding how the agent steps through the environment. The reward.py file contains information on the variety of reward types available for testing. agents.py contains information on the deterministic people that we created for our simulation.

gym-socialgame