2.7.1
The "Quantile of 0" printed out for the posterior mean summary is
now "2*min(Pr(>0),Pr(<0))". Also, previously, this quantity was
not doubled.
A bug in simulation for normal families have been fixed;
dispersion is now handled correctly.
2.7.0
A bug in the calculation of BIC has been fixed: the effective
number of parameters in the likelihood was not correctly
calculated. If the BIC is only used in selecting the number of
clusters, the qualitative results should not be affected (i.e.,
the selected cluster count should be the same), and similarly if
it is used to select the fixed effects to use in the model. (This
BIC should not, in general, be used to select the latent space
dimension or whether or not to include random effects.)
bic.ergmm() no longer prints out a warning with random actor
effects models.
bic.ergmm() now takes an additional argument, eff.obs, specifying
how to calculate the effective number of observations. The default
prior to this release has been the number of actors in the
network. The new default is the number of ties, as Handcock et
al. (2007) recommend.
latentnet no longer accesses unexported object in coda's
namespace, as it no longer needs to redefine coda's generics.
latentnet now exports the minimum necesary functions. Current
dependencies should not be affected, but if you would like some
function to be exported for your package, please let the
maintainer know.
Minor memory bugs have been fixed.
2.6.0
Normal model with identity link can now estimate the dispersion
(variance) parameter. (Before, it had to be specified.)
Improved compatibility with ergm 3.2, including removal of
mcmc.diagnostics() and other prototypes.
A number of bugs have been fixed, including some code warnings
from the latest versions of R.
Kirk Li and Li Wang have been added as contributors for making
many of these fixes.
Sampson's monks dataset included with latentnet now has some
additional attributes.
2.5.1
A bug in importing of ergm terms for undirected networks has been
fixed.
2.5.0
The deprecated term `latent` has been removed. Use `euclidean`
instead.
summary() and print() functions have somewhat more intelligent
defaults.
latentnet can now use ergm() terms as covariates in the model (and
the prior can still be specified). These should work seamlessly
for networks without self-loops. ergmm() terms to model self-loops
explicitly have been added.
Intercepts are handled a bit more intelligently.
2.4.5
A workaround for the overridden as.mcmc() generic has been added.
All examples lines are now shorter than 100 characters.
2.4.4
The implementation of Procrustes analysis has been reworked in two
ways: 1) latentnet no longer depends on the shapes library; and 2)
the C implementation of procrustes no longer uses the (deprecated)
EISPACK, replacing it with LAPACK.
Commonly used non-statistical functions have been extracted into a
new package, statnet.common, on which latentnet depends.
All contributors (as opposed to only the major ones) are now
listed in the DESCRIPTION file.
3D plots can now draw segments (for undirected graphs) and arrows
(for directed graphs, with package heplots) between vertices.
More examples and tests have been made conditional on environment
variable ENABLE_statnet_TESTS being set.
2.4.3
In order to make upgrading from the preview release automatic,
this version number has been skipped.
2.4.2
Latentnet is now compliant with R 2.15.0.
Latentnet no longer relies on a certain behavior of .C() that was
changed in the development version of R.
Many small updates, robustifications, and bug fixes have been
made. In particular, crashes and weird memory behavior with recent
versions of R and GCC have been fixed.
Author and maintainer e-mail addresses and affiliations have been
updated.
License specification has been clarified.
Latentnet now prints a warning when a matrix is passed on the LHS
of the model formula instead of a network.
Calculation of geodesic distances has been "outsourced" to the sna
package, which is now a dependency.
Fewer functions are unnecessarily masked.
Documentation has been cleaned up and updated. Examples and tests
have been shortened, speeding up R CMD check.
2.4.1
Latentnet now prints a warning when computing BICs for random
effects models, since we do not have a theoretical justification
for that.
A bug reported by Jorge González in the multithreaded code has
been fixed.
2.4.0
Removed the dependence on package 'mclust'. (The functionality
used has been reimplemented internally.) This means that
commercial use of the latent position clustering functionality
is now possible without the UW license. The reimplementation
requires 'mvtnorm' as a dependency.
Fixed a number of bugs reported by Soriani Nicola.
Modified the proposal tuner to propose those variables with high
autocorrelation more aggressively (at the expense of the
others). This should vastly improve mixing for complex models.
Package 'coda' is now a dependency rather than a suggested package.
Made the require()s for the suggested packages more informative.
2.3.0
Added an ability to fit bilinear latent space models of
Hoff (2005).
Bug fixes, including one reported by Soriani Nicola.
Documentation fixes.
2.2.3
Fixed a bug in handling of missing data, uncovered by Xiaoyue
"Maggie" Niu, and added a test for it.
Documentation fixes.
2.2.2
More documentation fixes.
2.2.1
Documentation fixes.
2.2.0
Ability to fit random effects (actor heterogeneity) models.
Removed ergmm.par class --- it has been rendered obsolete by
changes in R's handling of partial matching in [[ operator.
Many bug fixes and miscellaneous changes.