Context-Specific Likelihood Weighting

A sampling algorithm for Probabilistic Logic Programs.

Installation

This installation manual is tested on Ubuntu 18.04.4 LTS and 16.04 LTS.

1. Install dependencies

    $ sudo apt install build-essential pkg-config libgsl-dev

2. Install latest version of SWI-Prolog (version above 8.1.30 is required)

    $ sudo apt-add-repository ppa:swi-prolog/devel
    $ sudo apt-get update
    $ sudo apt-get install swi-prolog

3. Build

    $ cd Sampling
    $ make clean
    $ make all

Execution

4. Go to Examples folder

   $ cd ../Examples/

You will find one example of DC(B) program, namely, "alarm.pl". Alarm: (https://www.bnlearn.com/bnrepository/discrete-medium.html#alarm). Note that how tree-CPDs are written in the form of rules in DC(B) program.

5. Open an example in SWI-Prolog

   $ swipl -s alarm.pl

SWI-Prolog should now be opened without any error or warnings. First, set the number of samples.

   ?- set_sample_size(1000).

Second, turn off the debug mode (1 to turn on the debug mode).

   ?- set_debug(0).

Now query. The first argument is query, second is the list of evidence and P is the output probability

   ?- query(bp~=low, [lvfailure~=false, cvp~=normal, hr~=normal, expco2~=low, ventalv~=low, ventlung~=zero], P).