test-NestedKriging.R design/test sizes to avoid
a check timeout on slow CRAN workers
(e.g. r-devel-linux-x86_64-fedora-*, which exceeded the
45-minute test time limit under 1.1-0).New NestedKriging class: a divide-and-conquer
Gaussian process for large designs. The data are partitioned into
groups, one Kriging submodel is fitted per group with a
common prior, and predictions are aggregated with the optimal
nested-kriging aggregation ("NK", interpolating) or a
product-of-experts rule ("PoE", "gPoE",
"BCM", "rBCM").
New Vecchia approximated log-likelihood objective for large
designs: fit a Kriging model with
objective = "VLL(m)" (or "VLL", default
m = 30), costing O(n m^3) per evaluation instead of
O(n^3).
Kriging() / fit():
objective now also accepts "VLL" /
"VLL(m)", and regmodel now accepts
"quadratic".
Kriging() / fit(): the
noise argument has been moved to the last
position, for consistency with WarpKriging and the other
language bindings. Code that passes noise by name is
unaffected; positional calls that relied on noise being the
4th argument must be updated.