mikerabat / Mrai

// ################################################################### // #### This file is part of the mathematics library project, and is // #### offered under the licence agreement described on // #### http://www.mrsoft.org/ // #### // #### Copyright:(c) 2014, Michael R. . All rights reserved. // ###################################################################

This library contains classes for easy classifier design. For simple classifier design check out the TestApp - basically this is a unit test application -and see there which classifieres are available and how to use them. It's recommended to have base knowledge about the underlying algorithms. Quite a few algorithms have been designed with robustness in mind (robustness in terms of outlying pixels or occluded areas).

The package includes:

• Standard Fisher LDA classifier
• Robust (and Fast Robust) version of this classifier
• Incremental (and Robust) Fisher LDA classifier learning.
• Support Vector Machines (least squares and lagrangian learning)
• Naive Bayes
• Simple Decission stumps
• Radial basis function
• C4.5 Decission trees.
• K-means
• Ensemble classifiers: AdaBoost, Gentle Boost, Bagging
• Simple feed forward Neural Nets

On top of these classifiers there exists a few image database handling routines and an 1D, 2D Haar Feature extractor which is based on an integral image approach.

A testing application TestClassifier.dpr which shows the usage and performance of these classifiers on various tasks (e.g. face recognition).

Installation:

This library is an extension of the mrMath and mrImgUtils libraries and therefore depending on it!

You can check out the libraries from http://www.mrsoft.org/

or https://github.com/mikerabat/mrmath https://github.com/mikerabat/mrimgutil

First download both of these libraries and compile the included dpk files. Also add the directories to the library (and or) search paths!

Platforms:

This library was built mainly for the windows platform and is compatible with Delphi2007 and later versions.

Developer:

Rabatscher Michael contact via www.mrsoft.org

changelog:

Date 27.4.2020:

• speed up of the c4.5 implementation (thanks for the valuable input of N. Peladeau)
• Multithreaded c4.5
• Added some new features for c4.5: Stopping if a tree part has less then a certain minimum examples.

Date: 12.06.2017:

• Implemented new idea for weighted neural network learning (weight multiplication on backprop step)

Date: 16.03.2017:

• started change log
• Added quite a few confidence calculations: svm, lda, neural network, rbf, naive bayes, kmeans