logicGP

What is logicGP?

Introduction

A Framework for Literal Based Classification

At a glance

  • New hypothesis space for supervised learning based on literals
  • Hypotheses in the space may be represented as trees
  • Genetic Programming may be used to obtain the hypotheses

Classification


sepall sepalw petall petalw Class
5.1 3.5 1.4 0.2 Setosa
7.0 3.2 4.7 1.4 Versicolor
6.3 3.3 6.0 2.5 Virginica
… … … … …

Literals

  • Term is commonly used in Boolean algebra
  • Atomic formulas or their negations
  • Evaluating to true or false
  • Examples for classification:
    • sepal length > 6.0
    • sepal length ≤ 6.0
    • petal width ∈ [1.9; 2.5]

logicGP Classification Model

wseto wvers wvirg Condition
1.00 0.00 0.00 None below fulfilled
0.00 2.31 1.39 petal width ∈ [1.4;1.6]
0.00 5.56 0.00 sepal width ∈ [2.0;3.0] ∧ petal width ∈ [0.4;1.3]
0.00 0.00 7.00 petal length ∈ [5.2;6.9]
0.00 0.00 7.00 petal width ∈ [1.9;2.5]

Genetic Programming


Some Results

Implementation

Why C#?

  • Common choice for enterprise software
  • Superior performance in comparison to Python
  • More modern than Java

Existing Projects

  • ML.NET is Microsoft’s open-source machine learning (ML) framework for .NET
    • Necessary and helpful but difficult to expand
  • aima-csharp provides C# implementations for algorithms discussed in Artificial Intelligence - A Modern Approach
    • Incomplete and outdated
  • GeneticSharp is a Genetic Algorithm library
    • Focussed on Genetic Algorithms and not Genetic Programming
  • Genetica is a Genetic Programming library
    • Outdated

New Open Source Projects

  • Italbytz.ML provides extensions for ML.NET necessary for the implementation of custom algorithms
  • Italbytz.Ports.Algorithms.AI and Italbytz.Adapters.Algorithms.AI
    • Provide C# interfaces and implementations for algorithms discussed in Artificial Intelligence - A Modern Approach
    • Offers Genetic Programming library for the implementation of logicGP
    • Genetic and Evolutionary Computation library inspired by FrEAK planned

Genetic and Evolutionary Computation Library

FrEAK



Current Situation

  • Very pragmatic approach with focus on logicGP
  • No focus on UI
  • Generalizability and expandability was maintained
  • Greater use of the library in teaching planned
  • Further development in cooperation with external Open Source developers would be great

(1+1) EA

var onePlusOneEA = new EvolutionaryAlgorithm
{
    FitnessFunction = new OneMax(),
    SearchSpace = new BitString(),
    AlgorithmGraph = new OnePlusOneEAGraph()
};
onePlusOneEA.Initialization = new RandomInitialization(onePlusOneEA);
onePlusOneEA.StoppingCriteria =
[
    new GenerationStoppingCriterion(onePlusOneEA)
];
await onePlusOneEA.Run();
public class OnePlusOneEAGraph : OperatorGraph
{
    public OnePlusOneEAGraph()
    {
        Start = new Start();
        Finish = new Finish();
        var mutation = new StandardMutation();
        var selection = new CutSelection();
        Start.AddChildren(mutation, selection);
        mutation.AddChildren(selection);
        selection.AddChildren(Finish);
    }
}
% dotnet test --filter "Name=TestOnePlusOneEA"
Gen | Fit | Genotype
--- | --- | ----------------------------------------------------------------
  1 |  26 | 0011001000011010000100100110100110000000010001001011101110010111 
  2 |  26 | 0011001000011010000100100110100110000000010001001011101110010111 
  3 |  26 | 0011001000011010000100100110100110000000010001101011101010010111 
  4 |  26 | 0011001000011010000100100110100110000000010001101011101010010111 
  5 |  28 | 0011001000011010000100100110100110000100010001101011101110010111 
...