glmulti package changelog:
Version: 0.5-3 (2009-09-15)
Fixed: some issues in the coding of models
Version: 0.6-1 (2009-11-25)
fixed: some issues when using marginality and several covariate:covariate interactions.
glmulti now uses a S4 class ans S4 methods. It returns an object instead of writing a file.
Formulas are accepted to specify the candidate set of models.
Additional functions are provided to help representation of the results, model-averaging
and parallel computating (plot.glmulti, print.glmulti, summary.glmulti, coef.glmulti, predict.glmulti, consensus).
QAIC and QAICc are now builtin.
The genetic algorithm is now much faster.
Version: 0.6-2 (2009-12-1)
fixed: bug when using plot.glmulti with type="s"
fixed: startup of the java VM
fixed: minor bug when using method="r"
Version: 0.6-3 (2010-09-1)
added: weightable function to automatically generate a table of IC values and weights (thanks to Jarrett Byrnes)
added: out-of-the-box support for negative binomial glms (glm.nb)
fixed: bug when using predict and coef on lists of glmulti objects, and when feeding newdata to predict (thanks to Jarrett Byrnes)
Version: 0.6-4 (2010-12-1)
added: coef: confidence intervals are now returned for each estimate, in addition to the unconditional variance
fixed: coef: small problem when using Johnson's method to compute the unconditional variance
fixed: coef: small issue when the null model was in the confidence set: importance of the intercept was a bit underestimated
changed: the number of calls to java has been reduced by preparing several (parameter bunch) model formulas at a time to be fitted in R all at once.
Version: 0.7 (2011-01-20)
added: much faster method "l" for exhaustive screening with only covariates, using a branch-and-bound algorithm.
added: built-in support for mixed-models (lmer from lme4)
changed: all multi-model inference is now to be done with coef. Hence the glmulti objects no longer contain the codes and termweights fields. Summary no longer returns term weights. Plot type "s", and model coloring (highlight), is no longer supported.
improved: coef is more robust and presents terms sorted by importance.
fixed: some bugs arising when using exclusions with level=2.
fixed: bug arising when using quasi-models and QAIC/QAICc
Version: 0.8 (2011-04-01)
fixed: bug in coef yielding to small errors in the model-averaged estimates and inflated uncond. variances.
Version: 1.0 (2011-04-10)
First 'stable' release! Behavior has now been reasonably (still no warranty though!) checked for all provided functions.
New: coef() can now compute model averaged confidence intervals according to 3 methods: standard/Lukacs/Burnham
New: predict() can now return standard errors and CI of model-averaged predictions (via se.fit=T).
New: fitted model objects are now by default stored in the glmulti object, making coef() and predict() much faster.
New: plot now supports type="r", presenting diagnostics of the residuals for the best models.
Version: 1.0.1 (2011-04-10)
Added: Backward compatibility of coef(), predict() and summary() with earlier (<1.0) glmulti objects.
Changed: More consistent treatment of degrees of freedom in coef() and predict(). getfit() now returns degrees of freedoms as a third column.
Fixed: Idiosyncratic bugs in coef(), caused by R switching types of matrices.
Fixed: tendency to underestimate the unconditional variance of poorly suported effects in coef().
Fixed: problem in coef() when passing additional arguments to the fitting function.
Added: possibility to choose the mode of variance weighting in predict(), just like in coef().
Version: 1.0.2 (2011-07-15)
Added: Support for Cox prop hazards regression (coxph in package survival)
Added: Possibility to call glmulti on a list of fitted model objects, for flexibility and/or small candidate sets
Fixed: Behavior of coef and predict for some pathological cases: only one candidate model (well...), and handling of models with no estimated coefficient at all
Version: 1.0.3 (2011-11-01)
Restored: method "s" in plot.glmulti (for quick visualization of variable importances)
Added: Doc glmulti.pdf presenting the general functioning of glmulti, with the example of supporting lmer objects
(Thanks to Jurgis Sapijanskas)
Changed: aic is now the default criterion used, rather than aicc.
Version: 1.0.4 (2012-02-15)
Note: glmulti now depends on R 2.13
Changed: conforming to R 2.13. the generic function nobs is now used to get number of observations to compute information criteria (bic, aicc, qaicc)
Changed: conforming to R 2.13. the public functions saveRDS and readRDS are now used rather than their hidden counterparts (function write and consensus)
Fixed: some debugging messages were printed when running coef.glmulti
Added: illustrations on how to use glmulti with functions clm (ordinal) and multinom (nnet), in glmulti.pdf
Version 1.0.5/1.0.6 (2012-08-25)
Fixed: function consensus() did not order model objects properly, when using includeobjects=T
Fixed: slot "K" was sometimes not well aligned with models, ics and formulas
Version 1.0.7 (2013-04-10)
Changed: the estimated c-hat value to be used with QAIC/QAICC is now to be passed through option "glmulti-cvalue" rather than through the global variable "glmultiqaiccvalue" as it is used to be.
Fixed typo in qaic/qaicc (thanks to Marin Greenwood)