compute_H_W: Compute all the w_i blocks of H

In JimSkinner/spca: Structured PCA

Description

Compute all the w_i blocks of H

Usage

compute_H_W(X, WHat, muHat, sigSqHat, K)

Arguments

X	Data
WHat	Loadings matrix
K	Prior covariance matrix

Value

H_w_i

Examples

set.seed(1)
d=10; k=3; n=10
X = matrix(rnorm(n*d), ncol=d)
W = matrix(rnorm(d*k), ncol=k)
mu = rnorm(d)
sigSq = rnorm(1)^2
K = cov.SE(matrix(1:10, ncol=1), beta=log(c(1, 3)))

library(numDeriv)
library(Matrix)
Hw1.analytic = stpca:::compute_H_W(X, W, mu, sigSq, K)[[1]]
Hw1.numeric  = Matrix(numDeriv::hessian(function(w) {
  W_ = W
  W_[,1] = w
  -(log_likelihood(X, W_, mu, sigSq) + log_prior(K, W_, sigSq))
}, x=W[,1]))

stopifnot(all.equal(Hw1.analytic, Hw1.numeric))

JimSkinner/spca documentation built on May 7, 2019, 10:52 a.m.