SIGNIFICANT USER-VISIBLE CHANGES
eps
, the tolerance parameter determining which elements are set to zero has been increased.
New default is spam.eps=100*.Machine$double.eps
Operations from S4 Group Generic Functions Logic
and Arith
now all return a
regular matrix if one of the objects is a full spam
object.
SIGNIFICANT USER-VISIBLE CHANGES
nearest.dist()
has now a super fast Rcpp implementation by Annina Cincera.BUG FIXES
Arith
yielded wrong
results if the sparse matrix was full, see below. (Thanks to Ethan Heinzen.)INTERNAL CHANGES
Changes in CITATION
according to new format.
Eliminated kind
constructs in the Fortran code (addressing _R_CHECK_FORTRAN_KIND_DETAILS_
)
Added ...
argument to [t]crossprod.spam()
as by generic.
BUG FIXES
Logic
and Arith
yielded wrong
results if the sparse matrix was full. (Thanks to Ethan Heinzen.)BUG FIXES
## IGNORE_RDIFF_BEGIN
and eps=1e-15
in the help files to assure coherence
among different systems. SIGNIFICANT USER-VISIBLE CHANGES
rmvt.spam()
and rmvt()
sampling from a multivariate t-distribution.
Approach is similar to mvtnorm::rmvt()
.INTERNAL CHANGES
BUG FIXES
spam.eigen
.Prelease for version 2.9, only available on https://git.math.uzh.ch/reinhard.furrer/spam.
SIGNIFICANT USER-VISIBLE CHANGES
New function rgrf()
to sample a Gaussian random field (minimalistic replacement of
geoR::grf()
.
New implementation of the Matern covariance function for half-integer smoothness:
cov.mat12()
(wrapps to cov.exp()
) cov.mat32()
and cov.mat52()
for nu=1/2, 3/2, and
5/2, respectively. The function are factor 10 faster than cov.mat()
with corresponding
smoothness.
New function cov.finnmat()
using the parametrization sqrt(8smoothness)h/range for
as argument for the Bessel function.
All *likelihood*()
and mle*()
functions have now an argument cov.args=
which will
be passed to the covariance function. This creates a larger flexibility to fine-tune
covariance functions.
New function neg2likelihood.nomean()
, similar to neg2likelihood()
.
BUG FIXES
cov.nug()
, when range is smaller than eps
.SIGNIFICANT USER-VISIBLE CHANGES
Minor improvements in the help files. (2.8-1)
The function chol.spam()
has a much better verbose offering.
determinant.spam()
does not take any arguments for the Cholesky
factorization, if needed, pass via chol()
, i.e., det(chol(., args=))^2
.rmvnorm.conditional()
to sample from a multivariate conditional
normal random variable.rmvnorm()
and rmvnorm.spam()
have now arguments mean
and
sigma
(equivalent to mu
and Sigma
), to match mvtnorm::rmvnorm()
.article-jss
has been renamed jss10-figures-table
.
Deprecated demos article-jss-example1
, article-jss-example2
and
jss10-example2
have been eliminated.INTERNAL CHANGES
determinant.spam()
and chol.spam()
now use the common routine .spam.chol.basis()
.BUG FIXES
nnzcolindicies
was corrected (as in versions <=2.6).
The change in 2.7 was working well for huge matrices.
This addresses bug reported by Curtis Storlie, Martin Boer and Paul Eilers.determinant.spam()
now reflects what is calculated.determinant.spam()
should now work with 64-bit as well.chol.spam()
now has appropriate 64-bit
support for the subsequent iterations.eigen_approx
.
Debian package 2.7.0-2 contains the patch. Thanks to Sébastien Villemot.all.equal.spam(, scale=s)} now works for
length(s)>1, see PR#18272 for
base::all.equal.numeric()`, thanks to Michael Chirico. SIGNIFICANT USER-VISIBLE CHANGES
citeEntry
.covmat()
and nearest.dist()
.rmvnorm()
, cor.sph()
.INTERNAL CHANGES
covmat()
with elementary if()
clauses instead of ifelse()
, for stable memory usage.nearest.dist()
unleashed for spam64.BUG FIXES
SIGNIFICANT USER-VISIBLE CHANGES
print.spam()
is able to print non-zero entries of a spam matrix.gmult()
to multiply specific spam submatrices with different factors.INTERNAL CHANGES
BUG FIXES
SIGNIFICANT USER-VISIBLE CHANGES
spam_random()
and eigen.spam()
.eigen.spam()
for non-symmetric matrices is now stable.INTERNAL CHANGES
eigen.spam()
due to BLAS/LAPACK routines, which are used in Arnoldi iteration (ARPACK).BUG FIXES
INTERNAL CHANGES
NEW FEATURES
spam_random()
to create a random spam matrix.SIGNIFICANT USER-VISIBLE CHANGES
spam.options()
, spam.getOption()
are removed.validspamobject()
is now defunct.todo()
and spam.history()
are removed.summary.spam()
prints whether it is a 32 or 64-bit spam object.INTERNAL CHANGES
subset.rows.spam()
is now defunct.BUG FIXES
UScounties.ndorder
contained no-zeros on the diagonal. Now entire diagonal is zero.BUG FIXES
base::sample()
(http://developer.r-project.org/blosxom.cgi/R-devel/2019/02/26#n2019-02-26).SIGNIFICANT USER-VISIBLE CHANGES
cov.exp()
.BUG FIXES
det(spam(1))
bug fix & default of diag.spam(x)
in spam_diag()
removed.INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
BUG FIXES
INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
.Deprecated
.
'spam.options' -> 'options', 'spam.getOption' -> 'getOption'INTERNAL CHANGES
INTERNAL CHANGES
BUG FIXES
SIGNIFICANT USER-VISIBLE CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
Renaming of demos from JSS10 paper.
Continuing to implement 'spam_xxx' function names: Like 'spam_rdist' there is now 'spam_diag'.
INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
1.1.3.1 Suggested packages
approach.SIGNIFICANT USER-VISIBLE CHANGES
as.vector
for spam objects.spam_rdist
and 'spam_rdist.earthfor smooth use in
fields`.update(A, B)
without assignment has been eliminated.
This is one way to address the change in memory handling changes from R 3.0.2 to R 3.1.0.
There is a slight overhead in memory. If this causes problems, let me know.NEW FEATURES
diag
and eps
in nearest.dist
cause now an error.BUG FIXES
INTERNAL CHANGES
README
, inst/0LICENSE
.SIGNIFICANT USER-VISIBLE CHANGES
NEW FEATURES
BUG FIXES
INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
NEW FEATURES
BUG FIXES
INTERNAL CHANGES
NEW FEATURES
BUG FIXES
INTERNAL CHANGES
NEW FEATURES
BUG FIXES
INTERNAL CHANGES
BUG FIXES
SIGNIFICANT USER-VISIBLE CHANGES
NEW FEATURES
INTERNAL CHANGES
SIGNIFICANT USER-VISIBLE CHANGES
There is a generic conflict with 'backsolve' between spam and other packages (e.g., bdsmatrix). To avoid the issue, we use the standard generic implemented in 'methods' which requires an additional argument for version 0.29-0 (see also PR#14883). However to maintain backwards compatibility with packages that depend on spam, this was reverted in 0.29-1. Currently, this conflict is not properly solved. I propose to load 'spam' first then the other packages, followed by manually calling:
setMethod("backsolve","spam.chol.NgPeyton",backsolve.spam) setMethod("backsolve","spam",backsolve.spam)
Stay tuned...
Calls like:
mat <- diag.spam(4) diag(mat[-1, ]) <- 3 diag.spam(mat[ , -1]) <-2
now work. They are, however, somewhat inefficient. 'toeplitz.spam' is to be prefered. Pointed out by Florian Gerber.
The Gibbs sampler in the demo article-jss-example2 contains several bugs, pointed out by Steve Geinitz and Andrea Riebler. I'll post an updated sampler in a future release.
NEW FEATURES
INTERNAL CHANGES
BUG FIXES
NEW FEATURES
INTERNAL CHANGES
BUG FIXES
NEW FEATURES
INTERNAL CHANGES
BUG FIXES
NEW FEATURES
NEW FEATURES
NEW FEATURES
INTERNAL CHANGES
BUG FIXES
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.