Model for learning environment dynamics as logic relations, predicting future states and planning actions to reach reward.
pip install -r requirements.txt
Tweak options in model/constants.py
, such as:
DO_PRELOAD_HANDCRAFTED_*
- use handcrafted vectors instead of learnedVISUALIZE_*
- visualize stuffLEARNING_SOLVER
- one can use Gurobi to accelerate training, default is CBC
Run python3 run_agent.py