cpm.train.cv

K-fold cross-validation procedure to determine the optimal p-value threshold for feature selection

cpm.train.cv(data, outcome, p, nfolds = 5)

Arguments

data

An N x E matrix containing the vectorized edges; where N = number of subjects, E=number of edges

outcome

The outcome variable to predict

p

a vector of p-values to tested. If p is not specified, an automatically determined geometric sequence of p-values will be tested

nfolds

number of cross-validation folds. Set to 5 by default

Value

Returns a list object containing

• opt.pvals A vector containing the 2 optimal p-values, one each for the positive and negative network models

• results A P(number of p-values tested) x 2(positive and negative network models) matrix containing the predicted-actual correlations for each of the p-value selection thresholds in each network model.

• pvals A vector of the p-value threshold tested

Details

This function runs the cpm.train() using a range of p-values and determines the optimal p-values that maximize predicted-actual correlation in a K-fold cross-validation paradigm.

Examples

if (FALSE) {
model1.cv=cpm.train.cv(data=FC_data,outcome=dat_beh$age)
}