Added new family DirichletLSS for Dirichlet regression models (implemented by Michael Balzer)
Improved error messages for families with censored response.
Fixed minor bug of argument mismatch in internal S3 function get_qfun.
Updated ‘inst/CITATION’ file to reflect current guidance on how to cite in packages.
Fixed minor bug in cvrisk.
Fixed minor bug with out-of-bag risk for non-cyclical fitting.
Added www and a / after the URLs to make CRAN happy.
Changed http to https to make CRAN happy, hopefully finally.
Minor changes to make CRAN happy.
Initialize combined_risk in mboostLSS_fit to avoid writing
(and overwriting) the combined risk in the global environment.
Closes issue #49.
Fixed stabsel.mboostLSS for models fitted with FDboostLSS.
Closes issue #51.
Code in ‘man/families.Rd’ failed. Closes issue #53.
predint was not working when multiple variables were included
within a single baselearner.
Closes issue #55.
Fixed CRAN checks regarding all.equal by adding
check.environment = FALSE.
Export plot functions, streamlined interface of cvrisk and updated manuals. Closes issue #46.
Export predict.mboostLSS to make betaboost happy.
Updated reference for Thomas et al. Closes issues #47.
Added new fitting algorithm which can be used via the argument
method:
method = "cyclic" is the standard approach which was also
used previously.
method = "noncyclic" provides a new non-cyclical algorithm.
The "noncyclic" method allow faster cross-validation and better results
for stability selection (see below). For details see
Thomas, J., Mayr, A., Bischl, B., Schmid, M., Smith, A., and Hofner, B. (2017),
Gradient boosting for distributional regression - faster tuning and improved
variable selection via noncyclical updates.
Statistics and Computing. Online First.
DOI 10.1007/s11222-017-9754-6
(Preliminary version: https://arxiv.org/abs/1611.10171).
Stability selection (stabsel) implemented for gamboostLSS
models.
Models can now be fitted with zero steps (i.e., models containing only the offset). This change was mainly realized with the release of mboost 2.8-0. This closes issue #22.
The as.families interface to incorporate gamlss distributions
for gamboostLSS models was adapted to allow for other link functions.
Added option stabilization = "L2" to use the average L2 norm of
the negative gradient to make the updates for the different distribution
parameters comparable.
Fixed as.families("BB") and as.families("BI"), which is
actually an mboost family. Closes issue
#12.
Fixed as.families for 4-parametric "Mixed"-type families
(e.g. as.families("BEOI") and as.families("BEINF")).
Closes issue
#28.
Added Janek Thomas as new author.
Updated ‘inst/CITATION’ due to release of JSS tutorial paper and added new noncyclical fitting paper (also to manuals). The latter closes issue #27.
Updated URL in ‘DESCRIPTION’.
Adapt mboostLSS_fit() such that it can be called by FDboost::FDboostLSS().
Speed up computations by using response(fitted())
instead of fitted(, type = response) as the latter uses
predict(). Closes issue
#5.
Improved output of print.mboostLSS() and
summary.mboostLSS(). Closes issue
#7.
Added gamboostLSS_intern() to make FDboost
happy. Caution: Do not use this function.
Fixed ‘inst/CITATION’.
Fixed tests for stabilization.
Fixed bug in cvrisk.mboostLSS() which occurred if
families wasn't explicitly specified when fitting the
model. Closes
#9.
New interface: Stabilization of negative gradient is now
specified via families, e.g., GaussianLSS(stabilization =
"MAD"). Using options(gamboostLSS_stab_ngrad = TRUE) is
still supported for backward compatibility but discouraged.
Added weighted.median which is used in "MAD"
stabilization (see bugfixes).
Added preliminary version of the tutorial paper as vignette,
which is now available as vignette("gamboostLSS_Tutorial",
package = "gamboostLSS").
Added summary function (request
#2).
Added preliminary version of tutorial paper as reference.
Improve stored calls (closes
#3).
Use weighted.median to compute MAD when negative
gradients are stabilized.
Fixed handling of gamlss.dist families with type
= "Mixed".
Fixed a bug in as.families with truncated families
(which can be obtained using the package gamlss.tr).
The internal helper get_qfun is now a generic
function.
Adhere to CRAN policies regarding import of base packages (closes #1).
Changed plain text ‘NEWS’ to ‘inst/NEWS.Rd’
Changes in ‘inst/CITATION’ to make CRAN happy: Citations can now be extracted without the need to install the package.
Corrected india data set:
stunting ranged from -600 to 600 but should range from -6 to 6
Fixed unstated dependencies
Changed require to requireNamespace
and added gamlss.dist:: where necessary
Fixed bug in selected.mboostLSS
predict() returned a vector instead of a matrix
factors were coerced to characters in
predint()
added survival to suggests and load it in tests to reflect recent changes in mboost
removed the generic function risk() as this function is
now defined in mboost
check if response is correct for the family before
the offset is computed
re-added tests
changed \dontrun to \donttest in examples
added new function cvrisk for generic
cross-validation (gamboostLSS now requires mboost >=
2.2-3 for the generic function definition)
added a function to obtain cross-validation grids
(make_grid)
added new function as.families to include
gamlss families in gamboostLSS
added new families BetaLSS (implemented by Florian Wickler),
GammaLSS, GaussianLSS, ZIPoLSS, ZINBLSS
GaussianLSS is now the default for families
added new functions predint() and
plot(predint()) to obtain and plot marginal prediction
intervals
added stabilization factor for negative gradients (MAD):
turn on the stabilization by using
options(gamboostLSS_stab_ngrad = TRUE)
added new data sets india and india.bnd
lots of minor improvements in the interface and methods
improved manuals in various places
Speed of model fitting was dependent on iteration,
i.e. the algorithm got slower for larger mstop values
removed duplicated (and slightly altered) code from
mboost that was required to make gamboostLSS work with
earlier versions of mboost;
gamboostLSS now requires mboost >= 2.0-12
fixed potential problem that might occur if users specify different outcomes for different components
fixed bug in ‘tests/regtest-gamboostLSS.R’
changed the way we use environments
(to get rid of note: no visible binding)
updated dependencies
fixed bug in selected()
updated references
moved gamboostLSS to CRAN
some changes to make gamboostLSS work with mboost 2.0.x series
initial development version of gamboostLSS on R-forge