Rule Base

I. What It Is

Rule Base is used to store a set of abstract rules. Each rule has the form “under what conditions, what conclusion can be derived”, and is described in code by the Rule class:

body (premise)  →  head (conclusion)

For example (analogy only, not actual syntax):

  • If “a person is a student” and “this course is a required course”, then “this person must take this course”.

In this implementation:

  • RuleBase uniformly receives the initial rule list;
  • During initialization it uses RuleSpliter (DNF converter) to split compound rules into multiple sub-disjunctive rules, which facilitates subsequent reasoning.
  • Variables in rules are automatically renamed internally (_rename_rule_vars) to avoid variable name conflicts across multiple rules.

These processes are transparent to first-time users; you only need to write the rules in advance and hand them to the engine.

Different from Fact Base:

  • Fact Base changes incrementally during a single reasoning process (new facts will be added);
  • Rules in Rule Base are usually relatively fixed domain knowledge under an application scenario, and generally do not change frequently during reasoning.

Rule Base mainly does two things:

  1. Store all rules;
  2. Given a question Question, select a subset of possibly relevant rules to participate in this round of reasoning (initial_rule_base(question, topn=...)).

Therefore, from the user’s perspective, you can understand it as:

Rule Base = a domain expert’s “experience summary” and “business logic”.

II. Input Methods

2.1 String Form

WIP

2.2 Python Code

Entered via the Rule class.