Rdimtools (version 1.0.4)

oos.linear: Out-Of-Sample Prediction for Linear Methods

Description

Linear dimensionality reduction methods such as PCA, LPP, or ICA explicitly returns a matrix for mapping or projection. When we have new data, therefore, we can simply use the mapping provided. Inputs projection and trfinfo should be brought from original model you trained.

Usage

oos.linear(Xnew, projection, trfinfo)

Arguments

Xnew

an \((m\times p)\) matrix or data frame whose rows are observations. If a vector is given, it will be considered as an \((1\times p)\) matrix with single observation.

projection

a \((p\times ndim)\) projection matrix.

trfinfo

a list containing transformation information generated from manifold learning algorithms. See also aux.preprocess for more details.

Value

a named list containing

Ynew

an \((m\times ndim)\) matrix whose rows are embedded observations.

Examples

# NOT RUN {
## generate sample data and separate them
X = aux.gensamples(n=500)
set.seed(46556)
idxtest  = sample(1:500,20)        # 20% of data for testing
idxtrain = setdiff(1:500,idxtest)  # 80% of data for training

Xtrain = X[idxtrain,]
Xtest  = X[idxtest,]

## run PCA for train data
res_train = do.pca(Xtrain,ndim=2,preprocess="whiten")

## perform OOS.LINEAR on new dataset
## note that inputs should be from a given model you trained
model.projection = res_train$projection
model.trfinfo    = res_train$trfinfo
res_test  = oos.linear(Xtest, model.projection, model.trfinfo)

## let's compare via visualization
xx = c(-2,2) # range of axis 1 for compact visualization
yy = c(-2,2) # range of axis 2 for compact visualization
mm = "black=train / red=test data" # figure title
YY = res_test$Ynew  # out-of-sample projection for test data

opar <- par(no.readonly=TRUE)
plot(res_train$Y, type="p", xlim=xx, ylim=yy,
     main=mm, xlab="axis 1", ylab="axis 2")
points(YY[,1], YY[,2], lwd=3, col="red")
par(opar)
# }
# NOT RUN {
# }