Action Model Learning from Noisy Traces: a Probabilistic Approach

This repository contains the official code of the Noisy Offline Learning of Action Models (NOLAM) algorithm.

Installation

The following instructions have been tested on macOS Sonoma 14.5

1. Clone this repository:

 git clone https://github.com/LamannaLeonardo/OffLAM.git

2. Create a Python 3.9 virtual environment using conda (or venv):

 conda create -n nolam python=3.9

3. Activate the environment with conda (or venv):

 conda activate nolam

4. Install dependencies:

pip install numpy pandas matplotlib openpyxl

5. Check everything is correctly installed by running main.py script.

Execution

Running NOLAM

The NOLAM algorithm can be run for learning from traces with noisy states with a noise ratio varying from 0 to 1. To run NOLAM on all domains in "Analysis/Input traces/NOLAM/", and all noise ratios in [0, 1], run the main.py script.

Log and results

When you execute NOLAM, a new directory with all logs and results is created in the Results/NOLAM/noisy_states folder. For instance, when you run NOLAM for the first time, the logs and results are stored in the folder Results/NOLAM/noisy_states/run0. For each considered domain (e.g. blocksworld), a subdirectory is created (e.g. Results/NOLAM/noisy_states/run0/blocksworld), which consists of a log file and learned action model for each noise ratio in [0, 1]. In particular, each domain directory contain one subdirectory for every considered noise ratio (e.g. Analysis/Results/NOLAM/noisy_states/run0/blocksworld/0.2), where you can find three files:

1. log: contains some debugging information
2. model.pddl: the learned model
3. op_stats.json: for each operator and potential precondition/effect of the operator, we store the counting of transitions where the ground atom is true/false after/before executing an instantiation of the operator.

Finally, each run directory contains the detailed results over all domains in a pandas dataframe named nolam_results.xlsx.

NOLAM traces

For every domain and noise ratio in [0,1], the set of traces generated for evaluating NOLAM can be found in the directory Analysis/Input traces/NOLAM/.

Citations

If you find this repository useful, please consider citing the related paper.

@article{lamanna2024action,
  title={Lifted Action Models Learning from Partial Traces},
  author={Lamanna, Leonardo and Serafini, Luciano},
  booktitle={Proceedings of the International Conference on Automated Planning and Scheduling},
  volume={34},
  pages={342--350},
  year={2024}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.