django-rete - A Django-ORM based RETE-UL implementation

Overview

Implements a many-to-many database search based on the RETE-UL algorithm described by Robert Doorenbos in the paper "Production Matching for Large Learning Systems".

Allows creation of an arbitrary number of RETE networks, with the addition of productions and working memory elements (WME) in any SQL database supported by Django.

Note, this is not an attempt at implementing a complete production rule system. It's only the RETE component for performing a many-to-many query.

I decided to implement RETE-UL to clarify and test my own understanding of the algorithm. I chose to use Django as the underlying platform for a few reasons. First, because I was familiar with Django's ORM. And second, I was curious to evaluate an out-of-core database-backed RETE algorithm and was not aware of any public implementations.

This was started as a personal research project, and as such many features were not implemented or fully tested. Therefore, this project is intended more as a curiosity and is in no way appropriate for production use or as a replacement for traditional mature and robust in-memory RETE implementations (e.g. Clips).

A simple Django-based triple store (actually a quad store) is also included, as a means of storing working memory elements (WME) to feed into RETE networks.

Features

WME addition and removal
production addition and removal
retrieval of WME sets that match each production
right-unlinking
simple LHS conditional expressions (written in Python)

Installation

Install dependencies:

pip install django-uuidfield

Install the package:

python setup.py build
sudo python setup.py install

In your Django project's settings.py, add the 'rete' and 'triple' apps to your INSTALLED_APPS list.

You can run unittests with:

python setup.py tests

Usage

from rete.models import Rete, PNode
from triple.models import T

# Instantiate a new RETE network.
rete = Rete().save()

# Add a production to the network.
p1 = models.Production.get('rule1', [
    ['?','?x','on','?y'],
    ['?','?y','left-of','?z'],
    ['?','?z','color','red'],
])
rete.add_production(p1)

# Add WME to the network.
facts = [
    T('block1', 'on', 'block2'),
    T('block2', 'left-of', 'block3'),
    T('block3', 'color', 'red'),
]
for fact in facts:
    rete.add_wme(fact)

# Find a count of matched productions.
pnodes = list(PNode.objects.filter(_triggered=True).order_by('id'))
print len(list(pnodes[0].matches))

Todo