Codebase for plCoP, a Prolog Technology Reinforcement Learning Prover
Paper: https://arxiv.org/abs/2004.06997
git clone --recurse-submodules [email protected]:zsoltzombori/plcop.git
This project makes use of two external repositories:
- PySwip for Python-Prolog interaction: https://github.com/yuce/pyswip
- Hashtbl for prolog hash tables: https://github.com/gergo-/hashtbl
cd pyswip
pip install -e .
cd ..
foreign/xgb.so
However, depending on your system configuration, you may need to recompile it. See
foreign/INSTALL1
for an example how to do it. For the recompilation, make sure that the xgboost c library is in /tmp/:
cp foreign/libxgboost.so /tmp/
If you need to obtain another version of xgboost, this link can help:
https://xgboost.readthedocs.io/en/latest/build.html#building-the-shared-library
To run Monte Carlo Tree Search on a single tptp problem:
python montecarlo.py ini/plcop0.ini --problem_file <path_to_file>
To run a few iterations of learning on a directory of problems, see e.g.:
bash baseline.sh
