pca_OADP_proj: OADP projection

In privefl/bigutilsr: Utility Functions for Large-scale Data

OADP projection

Description

Online Augmentation, Decomposition, and Procrustes (OADP) projection of PC loadings onto some study data X.

Usage

pca_OADP_proj(X, loadings, sval)

pca_OADP_proj2(XV, X_norm, sval)

Arguments

X	Data to get PC loadings into.
loadings	PC loadings of the reference PCA to project.
sval	Singular values of the reference PCA (sqrt of the eigen values). Only the ncol(loadings) first ones will be used.
XV	X %*% loadings
X_norm	Vector of sums of squared rows (e.g. rowSums(X^2)).

Value

• pca_OADP_proj(): A list with the simple projection X %*% loadings and the projection based on OADP.

• pca_OADP_proj2(): The projection based on OADP only (a matrix of same size of XV).

Examples

X <- readRDS(system.file("testdata", "three-pops.rds", package = "bigutilsr"))
N <- 400; M <- ncol(X)
ind <- sample(nrow(X), N)

# Compute SVD using one part of samples
svd <- svds(X[ind, ], k = 5)
U <- sweep(svd$u, 2, svd$d, '*')
col <- 2:3
plot(U[, col])
points(cbind(0, 0), pch = 8, col = "green", cex = 2)

# Projecting other samples
proj <- pca_OADP_proj(X = X[-ind, ], loadings = svd$v, sval = svd$d)
points(proj$simple_proj[, col], col = "red", pch = 20)     # shrunk towards 0
points(proj$OADP_proj[, col], col = "blue", pch = 20)      # unshrunk

privefl/bigutilsr documentation built on Oct. 24, 2024, 1:45 p.m.