5.49.1
- Fixed bug in .generalizedF (Thanks to Dominic Edelmann and Fernando Castro Prado for pointing this out)

5.41.1
- Moved from KEGG.db to KEGGREST

5.37.1
- Fixed Y~1 Y~0 mixup in vignette
- Added fnctionality for competing risks (by Dominic Edelmann)

5.35.1
- Patched due to change in length.Surv
- Added ylim argument to covariates()

5.29.1
- Corrected error when calling features() when there is only one covariate.
- Corrected labeling of direction argument generated by "extract" in case of only positive or only negative associations

5.27.1
- Made gt return p-value=1 instead of a strange error if variance(response)=0 and model = linear (thanks to Renee Menezes for the bug report).
- Check and remove duplicated sets in inheritance() (thanks to Erik van Zwet for the bug report)
- repaired permutation test using too many permutations in logistic regression
- coerce multinomial model to logistic in case of two levels

5.23.3
- fixed a bug in the colours and color labels in the features plot in case of multinomial model (thanks to Lindsay Fitzgerald for the bug report)

5.23.1
- change default trace option to TRUE in gt.options
- fixed convergence criterion of .getC to avoid infinite runtime
- removed many require() calls. Added dependency on survival package.

5.13.1
- Fixed bug with negative weights (Thanks to Hugo Varet for the bug report)
- replaced turnListAround (internal) by a much faster version
- allowed parents with only a single child in inheritance
- changed as.character to deparse in name.response calculation in gt() to prevent bug in legend of features plot.
- fixed problem in covariates in the presence of zero weights leaves (thanks to Akdes Serin for the bug report)

5.11.6
- updated vignette

5.11.5
- removed plot import in NAMESPACE

5.11.4
- repaired bug in permutation test for survival response
- decreased the number of random permutations by 1.

5.11.3
- repaired bug in ordered factor coding
- removed all old version 4 stuff

5.11.2
- fixed bug giving too small p-values in test in logistic model without intercept
- changed the coding for ordered factors under the alternative to full split coding
- changed example on ordinal factors in vignette

5.11.1
- added plot argument to covariates()
- repaired extract() to also work with objects made with cluster=FALSE

5.9.3
- Minor change to documentation of gt-barplot

5.9.2
- fixed error message in case p-value is exactly 1

5.9.1
- repaired sorting and coloring of bars in covariates(..., cluster=FALSE, sort=TRUE)
- repaired colnames missing error if gt is called with a vector in alternative

5.7.4
- changed W==W[1] check in logistic model to max(abs(W-W[1]))<1e-15 to avoid rounding errors (Thanks to Florian Klinglmueller for the bug report).
- rewritten goodnessoffit.R function
- replaced gtPS, gtK, gtI, repdes, getPS, getK with gtPS, gtKS, gtLI, btensor, reparamZ, reweighZ in NAMESPACE/export 
- added sterms in NAMESPACE/exportMethods
- removed mboost and added gss, MASS, boot, rpart in DESCRIPTION/suggested
- updated Rd goodnessoffit file
- added sterms in Methods gt.object-class.Rd
- rewritten goodness-of-fit chapter in the vignette

5.7.3
- updated inheritance() to latest version of inheritance paper
- updated treatment of levels argument in gt: specifying levels coerces response to a factor
- updated treatment of levels argument in gt: specifying levels not in levels(response) now returns error
- fixed bug in cox model ignoring covariates

5.7.1
- added extract() method.

5.5.3
- changed calculation of permutation p-value according to paper Phipson and Smyth, SAGMB 2010.
- in gtConcept(), repeaired error if probe ids are entrez ids and annotation is org.Xx.eg.db package

5.5.2
- repaired imputation in case null covariate not estimable (Thanks to Mark van de Wiel for the bug report)
- removed zipData=yes from description

5.5.1
- changed homogeneous default in inheritance to true 
- repaired error: trying to get slot "weights" in inheritance with custom test
- repaired missing set names in inheritance with custom test
- added example for inheritance with custom test
- fixed error in mlogit in case mu numerically 1 or 0
- fixed zoom option in covariates for case alpha != 0.05


5.3.3
- changed homogeneous option of inheritance to the final version of the sequential rejection principle paper
- changed covariates() to use the homogeneous version of inheritance

5.3.2
- added the gtConcept function to the package

5.3.1
- rewritten .weed for better performance
- rewritten part of .genF to prevent numerical problems (non-convergence due to c==0). Thanks to Katsuhito Yasuno for the bug report.
- solved bug in .inherit associated with named subsets. Thanks to Jan Oosting for the bug report.
- Solved bug in covariates when cluster=FALSE and trace=TRUE. Thanks to Katsuhito Yasuno for the bug report.

5.1.6
- prevented warning in .getP if all lams<=0 (result p-value=0)
- prevented 0/0 problem with p.value exactly zero in inheritance
- prevented 0/0 problem with weight exactly zero in inheritance
- fixed imputation of missing values for cox model with no covariates
- more informative error message with missing annotation argument in gtGO, gtKEGG and gtBroad
- fixed covariates() for the case of a single covariate
- label "absolute correlation" in covariates plot abbreviated
- repaired multiple testing in gtGO, gtKEGG, GTBroad
- Repaired gtGO, gtKEGG, GTBroad in case of gt.options()$transpose is FALSE

5.1.5
- added Shaffer=TRUE to call to inheritance() from covariates()
- calculate ancestors from dendrogram rather than from sets to speed up covariates() in large sets
- optimized performance of covariates() and inheritance() for large data sets
- changed functions slot in gt.object from list to environment
- added zoom option to covariates()
- removed stop argument from inheritance() as it was not fully implemented anyway
- fixed gtGO() with multtest="focus" if ontology argument has length 1

5.1.4
- fixed input checking of formula object response
- readded option n=n from p.adjust call, but in an improved way
- made p.adjust case-insensitive for method specification
- repaired intercept-removal in a Cox model in case of factor covariates
- reimplemented .inherit to get the algorthm in line with the submitted paper

5.1.3
- covariates() with pdf option now removes empty subsets
- added option of negative indexing for gt.object
- removed n=n from p.adjust call, so p.adjust works better with NA p-values
- Worked around long variable names problem in covariates with large cluster tree
- added option that null input can be NULL, (equivalent to missing)
- made help.lines optional in covariates() and subjects(), defaulting to FALSE
- missing transpose leading to error in cox model with covariates fixed

5.1.2
- Added importFrom(AnnotationDbi, as.list) to NAMESPACE

5.1.1
- repaired bug in covariates() that gave wrong alias labels

4.99.12
- removed leftover debugging print command in .genF
- .calculateP() now uses F distribution rather than Imhof if possible
- repaired problem in covariates() in case of zero-variance covariates 

4.99.11
- updated Rd files (xrefs) to new Rcmd check standards.

4.99.10
- Repaired missing ] in v4_globaltest.Rd

4.99.9
- Added trimming of zero weights in gt()
- removed borders from covariates() and subjects() barplots
- repaired the object returned by covariates() if cluster=FALSE
- solved large sample numerical stability problem in .genF()
- made sure covariates() returns alias as well
- features() function added
- multtest() function replaced by p.adjust() method
- rewritten the introduction to the vignette to make it more acessible to non-statisticians
- subsets method added for gt.object

4.99.8
- updated Rd files (xrefs, S4 slot documentation) to new Rcmd check standards.
- fixed a terminal bug in covariates() in case of p-values exactly zero
- fixed a terminal bug in covariates() in case of a single covariate
- ancestor and offspring arguments added to inheritance()

4.99.7
- added a check for missing values in subsets argument of gt()

4.99.5
- fixed a bug in inheritance() sometimes causing a unended processing of the function.
- removed redundant levels in factor response in gt()
- transpose option fixed
- fixed the possibility of providing weights as a list of named vectors

4.99.4
- name conflict between new .weed() and deprecated .weed() resolved 
- GOstructure class, makeGOstructure(), and getFocus() placed back (deprecated) for the benefit of the GlobalAncova package
- line lengths of usage in .Rd files adjusted for reference manual
- option(warn.deprecated=FALSE) set in all deprecated .Rd files to avoind warnings during check

4.99.3
- globaltest(), gtGO() and gtKEGG() adapted to accomodate changes to annotation packages
- weights default changed in inheritance()
- arguments of inheritance() renamed

4.99.2
- minsize default in gtGO changed from 0 to 1
- focusLevel also proof against empty subsets

4.99.1
- Empty subset now gives NA p-value rather than error.

4.99.0
- Complete rewrite of the package: all function have been replaced, old ones deprecated
- New vignette
- New feature: complete support of formula-based calls to "gt" (was: globaltest)
- New feature: improved asymptotic null distribution for glms
- New feature: exact distribution for linear model
- New feature: inheritance method for multiple testing
- New feature: focus level method available outside GO context
- New feature: P-values, dendrogram and inheritance mutiple testing in "covariates" function (was: geneplot)
- New feature: P-values and dendrogram in "subjects" function (was: sampleplot)
- New feature: functions gtKEGG, gtGO, gtBroad for easier gene set testing.

4.15.1
- fixed bug in geneplot that deleted gene labels of single gene gene sets

4.13.1
- changed .pAsymptotic to only use true mixtures (c > 0) in order to avoid numerical problems in large data sets

4.11.3
- cleaned the dependencies in the description file: moved most packages from depends to imports field

- RG fixed up some issues with the vignette. It was using lookUp which means
annotate must be a depends; using mget instead removes that dependency

4.9.2 (2008-04-14)
- added a warning that method = "gamma" is deprecated.
- changed the makeGOstructure function to accomodate the removal of the GOALLENTREZID object from the GO package.
- updated the dependence of GO, KEGG and hu6800 to their .db versions.
- updated the sampling method so that it does not use the gamma z-scores anymore (Thanks to Roel Bakker for the bug report).


4.9.1 (2008-01-14)
- fixed a bug in the "[" method that arose when the permutations version of the test was used (Thanks to Remco Visser for the bug report).


