sunatthegilddotcom/42-expert-system

🧮 Backward chaining rule based system in Python. RPN, Tree resolver, Tree representation, logic rule system, prompt. A medium article is available in description.

42-expert-system

Python implementation of a backward chaining inference engine.

To understand how the resolver works, you can refer to my medium article here.

Presentation

The project receives an input file describing a set of rules, initial facts and queries. It must follows the following format:

# Example of input file

# Rules and symbols
C => E # C implies E
A + B + C => D # A and B and C implies D
A | B => C # A or B implies C
A + !B => F # A and not B implies F
C | !G => H # C or not G implies H
V ^ W => X # V xor W implies X
A + B => Y + Z # A and B implies Y and Z
E + F => !V # E and F implies not V
A + B <=> C # A and B if and only if C
A + B <=> !C # A and B if and only if not C

# Initial facts
= ABG

# Queries
?GVX

Getting started

Requirements

The project uses python3.7 and requires to install graphviz (for macs, do brew install graphviz).

We recommand you to use virtualenv to create an independant environment for the project.

pip3 install virtualenv
cd project-path/
virtualenv venv
source venv/bin/activate

Install python dependencies with:

python3 -m pip install -r requirements.txt

You can leave the virtual environment with the command deactivate.

How to use

The program works in two modes: shell and interactive. For more detail, python3 main.py -h.

usage: main.py [-h] [-m {shell,interactive}] [-g] [-r] [-i] [-s] [-v] input

Testing

python3 -m pytest -v

Implementation

Project files structure

README.md
/src
    Node.py # Most of the resolver functions are here
    Tree.py

Data structure

A tree represents a set of rules. It is composed of nodes divided in 3 types : ConnectorNode, AtomNode and NegativeNode.

ConnectorNode() # One of [+, |, ^, =>]
AtomNode() # Any uppercase letter, ex: A, B
NegativeNode() # Doesn't have an identity, it only represents the negative part 

The node class

All the node classes are child of a global Node class. Each node instance can have children. Child represent the left part of the equation. We can deduct that if a child is true, then its parent is also true.

The atom class

An AtomNode represents one fact. It's representation is an uppercase character. Each atom must be represented uniquely, meaning that we must never have any duplication. For example you will never find two A atoms.

AtomNode('A')

The connector class

A ConnectorNode is used to represent one operator in the ConnectorType set ( one of & | ^). The elements used for the calculation are saved in the operands property.

# Example: Create the (A + B) ConnectorNode

connector_ab = ConnectorNode(ConnectorType.AND)
node_a = AtomNode('A')
node_b = AtomNode('B')

connector_ab.append_operand(node_a)
connector_ab.append_operand(node_b)

The tree class

The Tree class keeps an AND operator connected to each of the Atoms in the rootproperty. So that we have one tree for the whole set of rules.

Equivalence

In the case of an equivalence, both nodes will be parent and child.

# Example: Creates the A <=> B
node_a = AtomNode('A')
node_b = AtomNode('B')
connector_imply = ConnectorNode(ConnectorType.IMPLY)
node_a.append_child(connector_imply)
connector_imply.append_child(node_b)
node_b.append_child(connector_imply)
connector_imply.append_child(node_a)

Handling priorities

We use the NPI notation to handle all operations priorities.

Credits

• @abbensid
• @jterrazz