yinsenm/extremeNN

a R implementation of extreme learning machine

[toc]

This is a draft implementation of Extreme Learning Machine in R. You can do bi-classification or multiple classification using elm_linear or elm_kernel.

Example

Linear ELM

1. Training Process

   r = elmTrain(X, Y, L = 2000, C = 300)

   Variable	Explanation
   X	the data matrix in training set
   Y	the labels from 1...N in testing set
   C	Penalty Parameter
   L	Number of hidden Layer

elmTrain will return a list r with the following:

|Result|Explanation|
|---|---|
|trainPred|the prediction of training set|
|confusion|the training confusion matrix|
|OutputWeight|OutputWeight to be used in `elmTest`|
|InputWeight| InputWeight to be used in `elmTest`|
|BiasofHiddenNeutrons|Biased Matrix to be used in `elmTest`|
|time|Time used|

2. Testing Process

   rr = elmTest(Xt, Yt, L = 2000, 
   	InputWeight = r$InputWeight, 
   	OutputWeight = r$OutputWeight, 
   	BiasofHiddenNeutrons = r$BiasofHiddenNeutrons)

   Variable	Explanation
   Xt	the data matrix in the testing set
   Yt	the label in the testing set
   L	Number of the hidden layer
   InputWeight	generated by elmTrain
   OutputWeight	generated by elmTrain
   BiasofHiddenNeutrons	generated by elmTrain

   elmTest will return a list r with the following:

   Result	Explanation
   testPred	the prediction of training set
   confusion	the testing confusion matrix
   time	Time used

The activate function used is sigmoid More activate functions to come!

Kernel ELM

1. Training Process

   r = elmTrain(X, Y, kernel_par = 1000, C = 200)

   Variable	Explanation
   X	the data matrix in training set
   Y	the labels from 1...N in testing set
   Kernel_par	Kernel Parameter used in rbfkernel
   C	Penalty parameter

elmTrain will return the following: |Result|Explanation| |---|---| |trainPred|the prediction of training set| |confusion|the training confusion matrix| |OutputWeight|OutputWeight to be used in elmTest| |time|Time used|

2. Testing Process

   rr = elmTest(X, Xt, Yt, kernel_para, r$OutputWeight)

   Variable	Explanation
   X	the data matrix in the training set
   Xt	the data matrix in the testing set
   Yt	the label in the testing set
   Kernel_par	Kernel Parameter used in rbfkernel
   OutputWeight	generated by elmTrain

elmTrain will return the following: |Result|Explanation| |---|---| |testPred|the prediction of testing set| |confusion|the testing confusion matrix| |time|Time used|

The kernel used is rbfkernel with more kernels to come!

Reference

1. extreme learning machine
2. Comparison among SVM, LS-SVM, PSVM, ELM