This repository contains source code and benchmarks for the IJCAI 2019 paper

Guillem Francès, Augusto B. Corrêa, Cedric Geissmann and Florian Pommerening.
"Generalized Potential Heuristics for Classical Planning."
Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI). 2019. 

The information and data contained in this repository should be enough to reproduce the experimental results in the paper. However, to stay updated with further developments in the codebase and explore other experiments not reported in the paper, we recommend checking the official code repository of the project: https://github.com/aibasel/basilisk

The code is developed in Python and requires both Python 3.5+ and a working installation of IBM CPLEX, which can be obtained free of charge for academic purposes, along with its Python API. The installation instructions and experiment results have been tested on an Ubuntu 16.04 machine with Python 3.5.2 and CPLEX 12.8.

Ubuntu Packages. The following Ubuntu packages are necessary:

sudo apt install python3-dev python3-pip 

Virtual Environment. For the installation of the Python code and dependencies, we strongly recommend to install everything on a Python3 virtual environment. The following would be a possible way of setting up a virtual environment dedicated to this project:

python3 -m venv frances-et-al-ijcai2019
source frances-et-al-ijcai2019/bin/activate
pip install --upgrade pip  # Upgrade Pip (optional, recommended)

The instructions that follow assume that both the python and pip commands refer to such a Python3 environment. It is important that you execute the instructions in the order given below.

CPLEX Python API. First install the CPLEX Python API: go to the directory /path/to/cplex/python in your CPLEX installation, and run

 python3 setup.py install

Tarski. Our project uses a slightly modified version of the Tarski Planning Problem Definition Module. For the sake of ease of use and reproducibility, we have included the codebase here. Go to folder tarski and run:

cd ~/frances-et-al-ijcai2019/tarski
 pip install .

Basilisk. Last, install this project's code itself. We recommend installing in development mode (-e). Go to the root folder of this distribution and run:

cd ~/frances-et-al-ijcai2019
pip install -e .

This should install all the required dependencies and leave the experiments ready to be executed.

Each individual experiment relies on a separate Python script that can be found in the experiments folder (see e.g. experiments/gripper.py for an example). The experiments shown in the paper should be reproducible by issuing the following commmands:

./experiments/gripper.py gripper_std_inc --all
./experiments/miconic.py miconic_1_incremental --all
./experiments/spanner.py spanner_1_incremental --all
./experiments/visitall.py problem03full_incremental --all