As part of my recent contract with I've been working on a project that is likely to have a large rulebase requiring frequent change. As a result we've decided to use a rules engine to manage them. We won't necessarily be allowing real time changes to rules, but we may well allow some changes between releases.
Jess was selected as it is already used within the organisation. It's not open source but meets the JSR-94 specification. It's based on the Rete algorithm and uses a CLIPS based language to define the rules. (CLIPS is a derivative of LISP so be ready for a lot of brackets...).
I have to say that it has taken me a while to get to grips with Jess as, for whatever reason, I ended up with the work of integrating it with our codebase. Trying to write the rules was incredibly difficult until I had a key realisation: Jess is not object oriented.
In fact the data (facts in Jess-speak) needs to be presented in a relational manner. In effect instead of references you need to create Primary and Foreign keys. Then you can create rules that navigate graphs of facts relatively easily. In effect you can start to leverage your experience with relational databases, the syntax is different but the thinking abou relationships is the same.
Once I had that realisation I produced a set of 'Fact' classes that present the Object Oriented data model in a relational manner. In order to bind the facts I define Jess templates with meaningful names from the 'Fact' classes and then use the Rete.addAsTemplate method to add the objects representing the data as Facts to Jess.
Jess is now much easier to work with.
No comments:
Post a Comment