2 include new sampled data example

[jakobdambon]

Using the spam::rmvnorm suite to sample SVC data, therefore replacing the old functions depending on package RandomFields.

Main changes:

• package contains a new sample data set called SVCdata, that has been sample from the RandomFields package. In the help file, I cite the RandomFields JSS article to give them the appropriate credit.
• package contains one sample function called sample_SVCdata.

[reinhardfurrer]

I had a quick look. seems reasonable. I will have a closer, detailed look once I am back in CH.

…

----- Original Message -----

From: "Jakob Dambon" ***@***.***> To: "jakobdambon/varycoef" ***@***.***> Cc: "reinhard" ***@***.***>, "Review requested" ***@***.***> Sent: Tuesday, 7 June, 2022 23:23:56 Subject: Re: [jakobdambon/varycoef] 2 include new sampled data example (PR #4) @jakobdambon requested your review on: jakobdambon/varycoef#4 2 include new sampled data example. -- Reply to this email directly or view it on GitHub: #4 (comment) You are receiving this because your review was requested. Message ID: ***@***.***>

--

______________________________________________________________________ University of Zurich ***@***.*** Department of Mathematics tel: +41-44-63-55843 and Institute of Computational Science office: Y27 J34 CH-8057 Zurich https://www.math.uzh.ch/furrer https://twitter.com/ReinhardFurrer

[reinhardfurrer]

Good to go. 2 cents:

For sample_SVCdata() should there be an X=NULL argument?
I tried to construct a nice figure (one-dim domain) to illustrate the additiveness of the SVC but with a rnorm() noise, one hardly sees anything.

Would it make sense to include the wrapper functions cov.mat32 and cov.mat52() in spam?
With these wrappers, you could simply construct a do.call() with any covariance function based on 3 parameters. Mainly aesthetics...

[jakobdambon]

I've added the X=NULL argument and pushed into master.

I don't understand your comment on the example / figure / additiveness of the SVCs. For a nice example, I use a code like this:

set.seed(123)
# SVC parameters
(df.pars <- data.frame(
   var = c(2, 1),
   scale = c(3, 1),
   mean = c(1, 2)))
# nugget standard deviation
tau <- 0.5

# sample locations
s <- sort(runif(500, min = 0, max = 10))
SVCdata <- sample_SVCdata(
  df.pars = df.pars, nugget.sd = tau, locs = s, cov.name = "mat32"
)

matplot(x = SVCdata$locs, y = SVCdata$beta, type = "l")

Could you provide an example?

For your final comment, I will put it in a backlog #5.

[reinhardfurrer]

I was thinking of:

set.seed(123)
# SVC parameters
(df.pars <- data.frame(
   var = c(2, 1),
   scale = c(3, 1),
   mean = c(1, 2)))
# nugget standard deviation
tau <- 0.5

# sample locations
s <- sort(runif(100, min = 0, max = 10))
X <- cbind(1, rep(c(0, 1, .5, 0, 1), times=c(24, 35, 5, 12, 24)))
SVCdata <- sample_SVCdata(
  df.pars = df.pars, nugget.sd = tau, locs = s, cov.name = "mat32", X = X
)
with(data = SVCdata, {
  plot( s, y, ylim = c(-1,6))
  lines(s, rowSums(X * beta), col = 2)
  lines(s, beta[,1], col = 3)
  lines(s, beta[,2], col = 4)
  lines(s, X[,2], col = 5)
})
legend("topright", legend = c('smooth fit', "first SVC", "second SVC", 
  "covariate second SVC"), col=c(2:5), bty = 'n', lty = 1)

It was helpful to understand the role of the covariate. I struggle with random covariates. Just a tiny detail, nothing to worry.