Skip to content

README.md

Latest commit

 

History

History
108 lines (81 loc) · 3.1 KB

README.md

File metadata and controls

108 lines (81 loc) · 3.1 KB

RobustFit.Core Library

Overview

RobustFit.Core is a .NET library for robust regression analysis, designed for fitting statistical models that are resistant to outliers. The library implements iteratively reweighted least squares (IRLS) methods with various robust loss functions, making it suitable for real-world datasets where anomalies may exist.

NuGet

Features

Regression Models

  • OLSRegressor: Standard Ordinary Least Squares regression
  • RobustRegressor: Robust regression models resistant to outliers

Loss Functions

  • HuberLoss: Huber loss function (default c=1.345) for a balance between efficiency and robustness
  • TukeyLoss: Tukey's bisquare loss function (default c=4.685) for complete resistance to outliers
  • ILossFunction: Interface for creating custom loss functions

Utility Functions

  • Linear Algebra: Matrix operations, vector manipulations, residual calculation
  • Statistical Measures: MAD (Median Absolute Deviation) calculations

Installation

NuGet Package Manager

Install-Package RobustFit.Core

.NET CLI

dotnet add package RobustFit.Core

Project Reference

<PackageReference Include="RobustFit.Core" Version="1.0.0" />

Usage Examples

Ordinary Least Squares Regression

using RobustFit.Core.Models;

// Create and fit OLS model
var ols = new OLSRegressor();
ols.Fit(X, y);

// Make predictions
double olsPrediction = ols.Predict(testPoint);

Robust Regression with Huber Loss

using RobustFit.Core.Models;
using RobustFit.Core.LossFunctions;

// Create loss function
var huberLoss = new HuberLoss(c: 1.345);

// Create and fit robust model
var robust = new RobustRegressor();
robust.Fit(X, y, huberLoss, maxIter: 100, tol: 1e-6);

// Make predictions
double robustPrediction = robust.Predict(testPoint);

Robust Regression with Tukey Loss

using RobustFit.Core.Models;
using RobustFit.Core.LossFunctions;

// Create loss function
var tukeyLoss = new TukeyLoss(c: 4.685);

// Create and fit robust model
var robust = new RobustRegressor();
robust.Fit(X, y, tukeyLoss, maxIter: 100, tol: 1e-6);

// Make predictions
double robustPrediction = robust.Predict(testPoint);

Custom Loss Functions

using RobustFit.Core.LossFunctions;

// Implement custom loss function
public class CustomLoss : ILossFunction
{
    public double Psi(double r) => /* Your influence function */;
    public double Weight(double r) => /* Your weighting function */;
    public double Loss(double r) => /* Your loss function */;
}

Performance Notes

  • Robust regression is typically more computationally intensive than OLS
  • Convergence is controlled by maxIter and tol parameters
  • For large datasets, consider preprocessing to remove obvious outliers first

Contributing

Contributions are welcome! Please feel free to submit a pull request or open an issue for any enhancements or bug fixes.

License

This project is licensed under the MIT License. See the LICENSE file for more details.