| Type: | Package |
| Title: | A New Reduced-Rank Linear Discriminant Analysis Method |
| Version: | 1.0 |
| Date: | 2015-11-04 |
| Author: | Yue S. Niu, Ning Hao, and Bin Dong |
| Maintainer: | Ning Hao <nhao@math.arizona.edu> |
| Depends: | R (≥ 3.1.1), MASS |
| Description: | A new reduced-rank LDA method which works for high dimensional multi-class data. |
| License: | GPL-2 |
| NeedsCompilation: | no |
| Packaged: | 2015-11-07 04:11:08 UTC; nhao |
| Repository: | CRAN |
| Date/Publication: | 2015-11-07 07:24:10 |
SPCALDA: A New Reduced-Rank Linear Discriminant Analysis Method
Description
A new reduced-rank LDA method which works for high dimensional multi-class data.
Details
| Package: | SPCALDA |
| Type: | Package |
| Version: | 1.0 |
| Date: | 2015-11-04 |
| License: | GPL-2 |
Author(s)
Yue S. Niu, Ning Hao and Bin Dong.
Maintainer: Ning Hao <nhao@math.arizona.edu>
A New Reduced-Rank Linear Discriminant Analysis Method
Description
A new reduced-rank LDA method which works for high dimensional multi-class data.
Usage
SPCALDA(X,Y,rho=exp(c((-2):6)),K=min(20,min(dim(X))), folds = NULL)
Arguments
X |
Input matrix, of dimension nobs x nvars; each row is an observation vector. |
Y |
Response variable for class label, of dimension nobs x 1. |
rho |
Tuning parameter. |
K |
The total number of principal components considered. |
folds |
Folds for cross-validation to select tuning parameter. |
Value
ob |
lda rule with top PCs |
tuneRotation |
Tuned rotaion matrix |
minerror |
Minimal training error |
rho |
tuned value of the parameter rho |
K |
tuned dimension, i.e., number of PCs |
Author(s)
Yue S. Niu, Ning Hao and Bin Dong
Examples
set.seed(2015)
n = 200;p = 500
X = matrix(rnorm(n*p),n,p)
mu=matrix(0,4,p)
mu[1,1:125]=0.4;mu[2,126:250]=0.4;mu[3,251:375]=0.4;mu[4,376:500]=0.4
Y = rep(1:4,50)
for (g in 1:4) {
index = which(Y == g)
n_g = length(index)
X[index,] = X[index,] + matrix(mu[g,],n_g,p,byrow=TRUE)
}
xtr = X[1:100,]; ytr=Y[1:100] #traning set
xte = X[101:200,]; yte =Y[101:200] # test set
folds = list(1:20,21:40,41:60,61:80,81:100)
spcaldaResult = SPCALDA(X=xtr,Y=ytr,rho=exp(c((-2):6)),K=20, folds = folds)
yhat = predict(spcaldaResult$ob,xte%*%spcaldaResult$tuneRotation)$class
error = sum(yhat != yte)