Screen_X: Screen features that are pure noise

In bingx1990/LOVE: Overlapping clustering based on latent factor models

Description

Screen_X finds features that are close to pure noise via k-fold cross-validation.

Usage

Screen_X(X, thresh_grid = NULL, nfolds = 10, nthresh = 50, max_prop = 0.5)

Arguments

X	A n by p data matrix.
thresh_grid	A numeric vector of thresholds. Default is NULL.
nfolds	The number of folds. Default is 10.
nthresh	Integer. The length of thresh_grid when thresh_grid is NULL.
max_prop	A numeric value between [0, 1] specifying the maximal proportional of pure noise features. Default is 0.5 meaning that at most 50% of features are pure noise.

Value

When only one value is provided in thresh_grid, Screen_X returns a vector of indices that are detected as pure noise. When either thresh_grid is NULL or multiple values are provided in thresh_grid, Screen_X returns a list including

• foldid The indices of observations used for cv.

• thresh_min The value of thresh_grid that has the minimum cv error.

• thresh_1se The largest value of thresh_grid such that the errors are wihtin one standard error of the minimum cv error.

• thresh_grid The used thresh_grid sequence.

• cv_mean The averaged cv errors.

• cv_sd The standard errors of cv errors.

• noise_ind a vector of indices that are detected as pure noise by using thresh_min.

See Also

KfoldCV_delta.

bingx1990/LOVE documentation built on Jan. 23, 2022, 1:30 a.m.