cv.pqc: Cross Validation for PQC

In obleeker/quant.pca: Principal Quantile Components Analysis

Description

Wrapper for pqcomp(): Select ranges for rank and lambda. Results are calculated on a random hold out on the supplied dataset.

Usage

cv.pqc(data, projDimRange, tau, lambdaRange, muEst = TRUE,
  epsilon = 1/log10(nrow(data)), shareNA = 0.1, iterTol = 200,
  convTol = NA, progBar = TRUE, doPar = TRUE, doSeq = FALSE)

Arguments

data	data matrix with rows as data entries and columns as variables
projDimRange	# range of desired principal components
tau	asymmetry parameter (b'w 0 and 1)
lambdaRange	regulization parameter
muEst	calculate a constant mu
epsilon	approxiamtion parameter
shareNA	size of holdout (0<shareNA<1)
iterTol	no. of max iterataions
convTol	set algorithm to stop if weigths did not change for convTol no. of consecutive iterations (deactivated for tau=0.5)
progBar	optional progressbar
doPar	use parallel backend (*nix systems recommended),
doSeq	run via sequential optimization
preOut	continues based on previous output (if doSeq is TRUE, the components of a previos PCA method are sufficient)

Value

list containing pqcomp() results

References

Madeleine Udell et al. (2016), "Generalized Low Rank Models".

obleeker/quant.pca documentation built on July 7, 2019, 12:41 a.m.