# cov.SE.d: Squared exponential covariance function derivatives wrt... In JimSkinner/spca: Structured PCA

## Description

Squared exponential covariance function derivatives wrt hyperparameters

## Usage

 `1` ```cov.SE.d(X, X2, beta, D = NA, ...) ```

## Examples

 ``` 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28``` ```point1 = matrix(rnorm(1), ncol=1) point2 = matrix(rnorm(1), ncol=1) beta = rnorm(2) # Logarithms of variance and length scale Ks.1point = cov.SE.d(point1, beta=beta) # Derivative wrt variance at zero distance should always be exp(beta[1]) stopifnot(all.equal(Ks.1point[[1]][1,1], exp(beta[1]))) # Derivative wrt lengthscale at zero distance should always be 0 stopifnot(all.equal(Ks.1point[[2]][1,1], 0)) # Identical tests with numerical gradient library(numDeriv) Ks.1point.num = grad(function(beta_) { cov.SE(point1, beta=beta_)[1,1] }, x=beta) stopifnot(all.equal(Ks.1point.num[1], exp(beta[1]))) stopifnot(all.equal(Ks.1point.num[2], 0)) Ks.2points = cov.SE.d(point1, point2, beta=beta) Ks.2points.num = grad(function(beta_) { as.numeric(cov.SE(point1, point2, beta=beta_)) }, x=beta) # Check numerical gradient equals analytic gradient stopifnot(all.equal(as.numeric(Ks.2points[[1]]), Ks.2points.num[1])) stopifnot(all.equal(as.numeric(Ks.2points[[2]]), Ks.2points.num[2])) ```

JimSkinner/spca documentation built on Aug. 19, 2018, 7 a.m.