rleap-project/batch-pddl-generator

Python library for creating sets of PDDL tasks

Batch PDDL Generator (BPG)

Specify PDDL generator parameters and their value ranges and let BPG generate PDDL tasks for you.

Installation

Create a virtual environment:

python3 -m venv --prompt sig .venv
source .venv/bin/activate
pip install -U pip wheel
pip install -r requirements.txt

Clone repo with PDDL generators, and build the generators:

git clone git@github.com:AI-Planning/pddl-generators.git
cd pddl-generators
./build_all

Usage

There are two ways in which this library can be used:

1. Generate the Cartesian product of instances over the given parameter values.

   For example, when you specify the following domain

    Domain(
        "tetris",
        "generator.py {rows} {block_type} {seed}",
        [
            get_int("rows", lower=4, upper=8, step_size=2),
            get_enum("block_type", ["1", "2", "3"], "1"),
        ],
    ),
   

   the command

    /generate-instances.py \
        --generators-dir <path/to/generators> \
        tetris /tmp/tasks
   

   will generate instances at /tmp/tasks for the following combination of rows and blocks:

    [(4, 1), (4, 2), (4, 3), (6, 1), (6, 2), (6, 3), (8, 1), (8, 2), (8, 3)]
   

2. Use SMAC to generate planning tasks that can be solved by a given planner within given resource limits.

    ./search-instances-for-planner.py \
        --generators-dir <path/to/generators> \
        --planner-time-limit 60 \
        tetris <path/to/singularity-planner.img>
   

Finding Duplicate Tasks

After generating the benchmark tasks, you might want to run the find-duplicate-instances.py script to detect duplicates.