choose_PCs.kurtosis: Selects the principle components (PCs) of sufficient kurtosis...

In muschellij2/clever: Calculate Leverage of Component Models

Description

First, the largest PCs which together explain 90 retained, and smaller components are removed. Each PC is detrended (this can be disabled). The kurtosis cutoff is then the 90 distribution of kurtosis for Normal data of the same length as the PCs; it is estimated by simulation or calculated from the theoretical asymptotic distribution if the PCs are long enough.

Usage

choose_PCs.kurtosis(svd, kurt_quantile = 0.9, detrend = TRUE, n_sim = 5000)

Arguments

svd	An SVD decomposition; i.e. a list containing u, d, and v.
kurt_quantile	PCs with kurtosis of at least this quantile are kept.
detrend	Should PCs be detrended before measuring kurtosis? Default is TRUE. Recommended if observations represent a time series.
n_sim	The number of simulation data to use for estimating the sampling distribution of kurtosis. Only used if a new simulation is performed. (If n<1000 and the quantile is 90 If n>1000, the theoretical asymptotic distribution is used instead.)

Value

The original indices of the PCs which were retained, in order of decreasing kurtosis.

muschellij2/clever documentation built on Sept. 26, 2020, 3:54 p.m.