LPMLN

An implementation to compute LPMLN programs using existing ASP and SDD solvers.

Introduction

LPMLN is a recent addition to probabilistic logic programming languages. Its main idea is to assign weights to stable models in a way similar to Markov Logic assigns weights to models. We present 3 implementations of LPMLN, namely lpmln-infer for inference, lpmln-learn for weight learning, and dtlpmln for DT-LPMLN inference.

Try with LPMLN Online

You can try LPMLN without installing it by using our online editor: http://reasoning.eas.asu.edu/lpmln-web/ . A simple user interface allows one to try LPMLN inference, LPMLN learning, and DT-LPMLN (for decision theory) without a hassle. Example projects (each consists of an LPMLN program and an evidence file) are also provided for one to start with.

Prerequisites

• System: Linux/Ubuntu
• Software: Anaconda
• Python: Python 3.6 or above

Installation

• [Optional] If your Python version is 3.5 or below, you may create a virtual environment in Anaconda

conda create -n env_lpmln python=3.7
conda activate env_lpmln

• Install Clingo and gcc

conda install -c potassco clingo
sudo apt install gcc

• cd to the folder of lpmln package, e.g., (Please modify the path to the lpmln folder accordingly.)

cd Downloads/lpmln

and execute the following commands

python setup.py
sudo ln -s $(pwd)/lpmln_infer.py /usr/local/bin/lpmln-infer
sudo ln -s $(pwd)/lpmln_learn.py /usr/local/bin/lpmln-learn
sudo ln -s $(pwd)/lpmln_dec.py /usr/local/bin/dtlpmln
chmod +x /usr/local/bin/*lpmln*
chmod +x $(pwd)/binSupport/*

Inference Example

Note that, if you created a virtual environment, e.g., env_lpmln, when you installed LPMLN, you need to activate this environment before you use the LPMLN package everytime you open a new terminal.

conda activate env_lpmln

Given the LPMLN program bird.lpmln, we can find the most probable stable model of it by executing the following command under the folder that contains this .lpmln file.

lpmln-infer bird.lpmln

Learning Example

Given the LPMLN program simple.lpmln and the observation file simple.obs containing observations (i.e., known facts) in the form of ASP constraints, we can learn the weight of the rules in smoke.lpmln such that the probability of the observations is maximized. Under the folder that contains this .lpmln file, we use the following command.

lpmln-learn simple.lpmln -o simple.obs

The learned weight are available in the file lpmln_learned.weight.