Parameter Seed Set

This repository contains code and supplementary material for Kokel et al. 2023.

Planning tasks succinctly represent labeled transition systems, with each ground action corresponding to a label. This granularity, however, is not necessary for solving planning tasks and can be harmful, especially for model-free methods. In order to apply such methods, the label sets are often manually reduced. In this work, we propose automating this manual process. We characterize a valid label reduction for classical planning tasks and propose an automated way of obtaining such valid reductions by leveraging lifted mutex groups. Our experiments show a significant reduction in the action label space size across a wide collection of planning domains. We demonstrate the benefit of our automated label reduction in two separate use cases: improved sample complexity of model-free reinforcement learning algorithms and speeding up successor generation in lifted planning.

Quick Start

This code has been tested on Ubuntu 20.04 and RHEL 8.5 for Python 3.8. It may not work on MacOS due to a known issue in one of the dependencies.

1. Clone this repo

git clone git@github.com:IBM/Parameter-Seed-Set.git
conda create -n pss python=3.8
conda activate pss
cd Parameter-Seed-Set
pip install -r requirement.txt
pip install -e .

2. Setup dependencies

This code makes system calls to the following libraries.

• CPDDL
• Forbid Iterative Planner

❗ If you already have CPDDL or a PDDL-based Planner, skip to step 3.

Build CPDDL. Following packages are required for successful build unzip automake autotools-dev

```
sudo apt-get install unzip automake autotools-dev
git submodule update --init --recursive -- ./dependencies/cpddl
cd ./dependencies/cpddl
./scripts/build.sh
cd ../..
```

Build ForbidIterative planner

```
git submodule update --init --recursive -- ./dependencies/forbiditerative
cd ./dependencies/forbiditerative
python ./build.py
cd ../..
```

3. Set environment variables

Skip this part if the CPDDL and Planner were installed as part of step 2.

Otherwise, configure your respective path to pddl-lifted-mgroups and fast-downward.py by declaring following environment variables.

# For CPDDL Lifted Mutex Groups
export CPDDL_LMGS_PATH=./dependencies/cpddl/bin/pddl-lifted-mgroups
# For Fast Downward planner
export FAST_DOWNWARD_PATH=./dependencies/forbiditerative/fast-downward.py

4. Run the runner.py to find the parameter seed set and compare the action spaces.

$ python ./runner.py \
--domain-file ~/downward-benchmark/blocks.pddl \
--problem-dir ~/downward-benchmark/blocks \
--use-grounding

Code organization

Repository
├── README.md                       # this file
├── LICENSE                        #  license file
├── pss/                           # contains core code
|  ├── util/                # utility files interfacing dependencies
|  ├── evaluation.py           # evaluation code
|  └── parameter_seed_set.py   # core formulation and solution.       
├── runner.py            # main runner file    
├── unittests/           # unit test cases
├── dependencies/        # git submodules
└── scripts/             # scripts to replicate empirical results

Citation

If you build on this code or the ideas of this paper, please use the following citation.

@inproceedings{KokelLKSS23,
 	title={Action Space Reduction for Planning Domains},
 	journal={IJCAI},
 	author={Kokel, Harsha and Lee, Junkyu and Katz, Michael and Srinivas, Kavitha and Sohrabi, Shirin},
 	year={2023}
}

LICENSE

This code is licensed under the Eclipse Public License, Version 1.0 (EPL-1.0).