I have recently fallen in love with the Rete algorithm initially developed from Charles Forgy.
After having implemented it (with my own personal vision) in Elixir and having it working for some basic cases, I would like to develop this further and have it ready for production for companies that want to use Rete as part of their system. Maybe even implement a DSL so that it is easier to use it.
I would like to know if anyone is keen to help me develop and design more feature for this interesting AI project in Elixir (there isn’t any open source expert system built in Elixir, that I am aware of).
Also, would be great to know if anyone may need a rule engine built with Elixir in their system so I can help you implement it and see what extra feature would be needed.
So any feedback and suggestion is very welcome. This is the source code for now https://github.com/lorenzosinisi/retex (feel free to open issues and PRs on the Github repo)
This is the paper from Forgy http://www.csl.sri.com/users/mwfong/Technical/RETE%20Match%20Algorithm%20-%20Forgy%20OCR.pdf
And this is YouTube video explanation about Rete https://www.youtube.com/watch?v=CmxHPJTAF3I and how it can be used in the real world.
Note: I am not an expert of AI so please bear with me if the initial implementation is not complete or completely correct, I just needed to “make it work” first in order to “make it beautiful” (and I need help for that)
Awesome! I am completely new to Elixir but have played around with the Python CLIPS libraries. I’m building a product that has a domain ontology encoded with Postgres LTREE as a network of DAG taxonomies with a custom triple store schema. There are trees for normal taxa stuff like topics, but also for procedures of rules. It needs breadth-fist search to match potential rules against iterative, mostly synchronous user input sessions, e.g. checking a fact pattern that emerges from interviews against laws.
From doing a couple throwaway prototypes I have become less and less convinced that a “real” graph (lib or db) is needed. Basic trees are fine because I can’t think of cases where multiple roots or cycles must exist, or even if they did, why they must be handled in the actual traversal algorithm if I can just use SQL relations and code to stitch together many branches when initializing a Rete network (not an expert on that so could be missing something). The attraction of Rete for me is doing breadth-first search for rule/action nodes and not having to think about pruning dead-end paths during execution. Any other theoretical capability feels like conceptual overhead, but I can appreciate the appeal
Since this is the only pure Elixir Rete lib I could find, I hope to give it a spin soon.
Have you looked at https://github.com/afiniate/seresye as well? They seem to just use tuples, which would seem to be quite nice fit with pattern matching. It is also the approach taken by some projects in other languages like https://github.com/project-flogo/rules.
Hope to see more people on this!
I previously worked on a horrendously complex promotions engine (for a retail system), and I’m interested in how a rules engine might solve this problem.
How far is this from being “production ready”?