fast.SS: Compressed spectral screening

In fastCorrDiff: Fast Differential Correlation Matrix Screening

Description

Compressed spectral screening for high-dimensional differential correlation matrices.

Usage

fast.SS(X1, X2, rho, K, K.seq = FALSE, sv = FALSE, tune = FALSE)

Arguments

X1	Matrix of the first group of observations. Each row is variable and each column is an observation. Note that this is different from the usual way of statistical data matrices, because usually number of variables is much larger than the number of observations in this case.
X2	Matrix of the second group of observations. Each row is variable and each column is an observation. Note that this is different from the usual way of statistical data matrices, because usually number of variables is much larger than the number of observations in this case.
rho	The sampling proportion for compressed screening.
K	The rank K or the maximum rank K to calculate the spectral scores
K.seq	logical value. If TRUE, then it returns the spectral screening scores for all rank from 1 to K. If FALSE (default), it returns the rank K result
sv	logical value. If TRUE, the square roots of singular values will be used in the score calculation.
tune	logical value. If TRUE, a cross-validation procedure will be called to estimate the the rank K by evaluating imputation errors of ten percent of additional sampled entries.

Details

This function can be used to tune the SS function as well. Also, the spectral projection method of Cai et al (2017) is also implemented as part of the function. So, it will return the scores by both compressed version of spectral screening and spectral projection. Notice that if one wants to analyze the Spearman's correlation, the input objects X1 and X2 should be the rank matrices instead of raw data matrices.

Value

SS	The result of the SS function on the incomplete correlation matrix
Cai	The result of the spectral projection of Cai et al (2017) on the incomplete correlation matrix
tune.error	The imputation error on hold-out entries
tune.time	Timing for the additional tuning step.

Author(s)

Tianxi Li and Xiwei Tang Maintainer: Tianxi Li [e-mail]

References

Tianxi Li, Xiwei Tang, and Ajay Chatrath. Compressed spectral screening for large-scale differential correlation analysis with application in selecting Glioblastoma gene modules. arXiv preprint arXiv:2111.03721, 2021.

See Also

SS

Examples

X1 <- matrix(rnorm(200),10,20)
X2 <- matrix(rnorm(200),10,20)

fit <- fast.SS(X1, X2, 0.3, K=2) 
fit$SS$score

fastCorrDiff documentation built on Nov. 18, 2021, 5:07 p.m.