README.md

Calypte

Unstable, under development.

Small embeddable, flexible rule-based engine framework. To be really embeddable, it uses pluggable architecture and concept of rule lifecycle. Users of rule engine will need to configure and plug and change behaviour via callbacks.

Example would be, how to implement persistence. Rule engine works in memory, but you can plug different behaviours, which can implement callback after_execute and persist changes. With this approach you can have different behaviours, like: SaveAfterExec, SaveBySideEffect and different storages (DGraph, Postgresql).

Via callback in lifecycle load_data the data will be loaded and should be updated, if there any change available in extern source (so it interact with a source of truth via loading data and actions and loading data back, so without direct persistence at all).

The rule language will have a pluggable interface for functions, which can be used to implement functions for the applied domain, and meta information, which can be accessed by lifecycle modules t o modify behaviour of rules based on this meta information.

Little bit complexer example: by implementing a function, this function can set meta information ( for example side_effect: true), and if a rule using this function, than module SaveBySideEffect can persist information if any side effect functions were used.

The concept of rule evaluation and execution life cycle and user defined functions should allow to embedd rule processing to different systems, storages and use cases.

The intern representation of data is a graph (candidate for it: libgraph).

Inference algorythm, which will be used is RETE (or inspired by RETE), so that rule condition graph is build, where the same conditions from different rules grouped togehter, so that we can check if a condition doesn't eval to true, than we can ignore all rules, which have this condition in one go.

Using forward chaining algorythm there is, because rule-based systems (in my personal opinion) should support dynamic situations and forward chaining is better suited for dynamic situations, where data changes and only subpart should be evaluated to produce new inferences. Backwards chaining shines more in static query-like analytic questions, where my goal to support long running dynamic inference cases first. But there are exstensions for RETE, which uses rete network for backwards chaining, so it can be extended and explored in a future.

Installation

If available in Hex, the package can be installed by adding calypte to your list of dependencies in mix.exs:

def deps do
  [
    {:calypte, "~> 0.1.0"}
  ]
end

Documentation can be generated with ExDoc and published on HexDocs. Once published, the docs can be found at https://hexdocs.pm/calypte.

Language

Goals to have compact language (exstensible via meta information and domain functions), which will looks like this (may be changed):

@if
  $father isa Person
    gender == "male"

  $child isa Person ($father has:child $child)
    age < 18

@then
  ...

Very initial Roadmap

• language
  • int - signed 64 bit integer
  • float/decimal - float or decimal
  • string - string
  • bool - boolean
  • datetime - datetime
  • list - list
  • meta information
  • type definitions
  • relationship matching
• rule compiler
• graph of interconnected nodes
  • node representation
  • graph traversing
• processing rule matching network (inspired by RETE)
  • building condition graph
  • propagate nodes to rules graph
• life cycle