This is for implementations of Subgoal-based Reward Shaping.
This implementation requires external packages, and you first install the packages by pip or other package management tools.
pip install -r requirements
You can run a learning agent in each domain.
python -m fourroom --config=fourroom/in/configs/fourooms.ini
python -m pinball --config=pinball/in/configs/pinball.ini
mpiexec -n 8 python -m picknplace --config=picknplace/in/configs/picknplace.ini
We place a config file per algorithm and label the file the name of algorithm like sarsa
.
If you want to try LEARNED with HUMAN, you selectlearned_human
.
The program for analysis is here.