predict.rda {rda}

RDA Prediction Function

Description

A function that predicts the class labels for new samples using RDA.

Usage

## S3 method for class 'rda'
predict(object, x, y, xnew, prior, alpha, delta,
            type=c("class", "posterior", "nonzero"),
            trace=FALSE, ...)

Arguments

object

An rda fit object obtained from the function rda.

x

The training data matrix as used in the 'fit' object.

y

The class labels for the columns of 'x' as used in the 'fit' object.

xnew

The new data matrix used to predict the class labels of the new samples. Must be a numerical matrix with rows corresponding to variables and columns corresponding to the samples. The number of rows must be the same as 'x'.

prior

A numerical vector that gives the prior proportion of each class. By default, it is set to the fit component from the training step unless users want to specify a new one for prediction.

alpha

A particular regularization value for alpha. Often, this is the optimal alpha value obtained from the cross-validation step. But it could be any other value that users set. A vector of values is also acceptable. If missing, the function will use the default values from the fit component.

delta

A particular threshold value for delta. Often, this is the optimal delta value obtained from the cross-validation step. But it could be any other value that users set. A vector of values is also acceptable. If missing, the function will use the default values from the fit component.

type

A character string specifying which type of prediction is desired. If 'class', then the predicted class labels are returned; if 'posterior', then the predicted posterior probabilities for each sample belonging to a class are returned; if 'nonzero', then the indicators of shrunken genes are returned. 'class' is the default value.

trace

A logical flag indicating whether the intermediate steps should be printed.

...

Additional arguments for generic predict.

Details

predict.rda does various predictions on the new test samples based on fit from the training samples.

Value

If option "type='class'", the function will return the predicted class labels for the new test samples. The format is a 3-dim array. The first index corresponds to the alpha value(s) while the second index corresponds to the delta value(s). The last index is the predicted labels for the new samples. A reduced-dimensional array is possible if the length of alpha or delta is 1.

If option "type='posterior'", the function will return the predicted posterior probabilities of the new test samples belonging to different classes. The format is a 4-dim array. The first index corresponds to the alpha value(s) while the second index corresponds to the delta value(s). The third index represents the samples in 'xnew'. The last index is the class labels. A reduced-dimensional array is possible if the length of alpha or delta is 1.

If option "type='nonzero'", the function will return a 3-dim indicator array of the shrunken genes by RDA with 3 indices corresponding to alpha, delta and the indices of the genes respectively. A reduced-dimensional array is possible if the length of alpha or delta is 1.

Author(s)

Yaqian Guo, Trevor Hastie and Robert Tibshirani

References

Y. Guo, T. Hastie, R. Tibshirani, (2006). Regularized linear discriminant analysis and its application in microarrays, Biostatistics 8 pp. 86–100. doi:10.1093/biostatistics/kxj035.

See Also

Also see rda and rda.cv.

Examples

data(colon)
colon.x <- t(colon.x)

## divide the data set into a training set and a test
## set using a ratio of 2:1.
tr.index <- sample(1:62, 40)
fit <- rda(colon.x[, tr.index], colon.y[tr.index])

## predict the class labels of the test set at alpha=0.1
## and delta=0.5
ynew <- predict(fit, x=colon.x[, tr.index], y=colon.y[tr.index], 
                xnew=colon.x[, -tr.index], alpha=0.1, delta=0.5)

## calculate the prediction error
sum(ynew != colon.y[-tr.index])