The function that does RDA analysis on high dimensional data, e.g., microarray expression data.
rda(x, y, xnew=NULL, ynew=NULL, prior=table(y)/length(y),
alpha=seq(0, 0.99, len=10), delta=seq(0, 3, len=10),
regularization="S", genelist=FALSE, trace=FALSE)The training data set. It must be a numerical matrix. The columns are sample observations and the rows are variables. For example, in the microarray settings, "x" is the gene expression matrix with the columns corresponding to the arrays while the rows corresponding to the genes.
The class labels of the training samples (columns) in 'x', which must be consecutive integers starting from 1.
The test data matrix. It has the same structure as 'x'. The columns are samples and the rows are variables.
The class labels of the test samples. Same requirement as for 'y'.
A numerical vector that gives the prior proportion of each class. Its length is equal to the number of classes. If not supplied, it is set to the sample proportions by default.
A numerical vector of the regularization values for alpha. A single value is allowed. If not supplied, the default one will be used.
A numerical vector of the threshold values for delta. A single value is allowed. If not supplied, the default one will be used.
Define which regularization method to use. 'S' stands for regularization on covariance; 'R' stands for regularization on correlation. 'S' is the default option.
A logical flag. If 'TRUE', then the function will return an array of indices indicating the genes remained for each (alpha, delta) combination. By default, this is set to 'FALSE'.
A logical flag. If 'TRUE', then the intermediate computation steps will be displayed. Caution: this would lead to a very long output display. By default, this is set to 'FALSE'.
The function will return an 'rda' object with the following list of components:
The vector of the regularization values for alpha used in the function.
The vector of the threshold values for delta used in the function.
The vector of the prior proportion of each class used in the function.
The training error matrix. The rows correspond to the alpha values while the columns correspond to the delta values.
A 3-dim array giving the predicted class labels of 'y'. The first index corresponds to the alpha values while the second index corresponds to the delta values. The third index is the predicted class labels for the corresponding samples. However, when the length of alpha or delta is 1, this could be a 2-dim matrix or even a 1-dim vector.
The matrix of the number of shrunken genes. The rows correspond to the alpha values while the columns correspond to the delta values.
The group centroids matrix. It has the same number of rows as 'x' and the number of columns is the total number of classes. Each column is the centroids vector of the samples within that class.
A single vector giving the grand mean vector of all the samples in the 'x' matrix.
A 3-dim array of the predicted class labels for the columns of 'xnew' if 'xnew' is provided. The first index corresponds to the alpha values while the second index corresponds to the delta values. The third index is the predicted class labels for the corresponding samples. However, when the length of alpha or delta is 1, this can be a 2-dim matrix or even a 1-dim vector.
A 4-dim array giving the posterior probabilities of each column of 'xnew' belonging to a class if 'xnew' is provided. The first index corresponds to the alpha values while the second index corresponds to the delta values. The third index is the corresponding columns in 'xnew'. The last index corresponds to different classes. However, an array of reduced dimensions may be produced if any of these four indices has length of 1.
The test error matrix if both xnew and
ynew are supplied. The rows correspond to the alpha values
while the columns correspond to the delta values.
A 3-dim array giving the indicator whether a gene is shrunken or not for a particular (alpha, delta) if "genelist" option is 'TRUE'. '0' means that gene is shrunken while '1' otherwise. The first two indices correspond to alpha and delta. A reduced-dimensional array is possible if either alpha or delta is of length 1.
The type of regularization used in calculation.
rda does RDA analysis on high dimensional data. This is the main
function of the package.
Guo, Y. et al. (2004) Regularized Discriminant Analysis and Its Application in Microarrays, Technical Report, Department of Statistics, Stanford University.
rda.cv and predict.rda.
# NOT RUN {
data(colon)
colon.x <- t(colon.x)
fit <- rda(colon.x, colon.y)
# }
Run the code above in your browser using DataLab