version 1.9.12: (2012-09-05)

- added function newCountDataSetFromHTSeqCount

version 1.9.7: (2012-05-25)

- fixed handling of rank-deficient observations now also in estimateDispersions

version 1.9.6:

- fixed that varianceStabilizingTransformation was missing in exports

version 1.9.5:

- added function plotDispEsts

version 1.9.4: (2012-04-24)

- fixed handling of rank-deficient observations

version 1.9.2: (2012-04-05)

- fixed bug in pooled-CR dispersion estimation

version 1.9.1: (2012-04-01)

- Added a new function 'varianceStabilizingTransformation'

version 1.7.10:  (2011-03-26)

- Added pooled-CR method.
- Fixed error message for insufficient replication.

version 1.7.9:  (2011-03-16)

- Fixed error in vignette.
- Silenced warnings from nbinomFitGLM about convergence; this information is given in the result anyway.

version 1.7.8:  (2011-03-15)

- The VST is now roughly equal to log2 for local dispersion fits, too.

version 1.7.7:  (2011-03-15)

- changed the variance stabilizing transformation once more; this time it is hopefully
  correct. The derivation of the formula is now documented in the new file vst.pdf.

version 1.7.6:

- fixed regression bug in getVarianceStabilizedData: the formula for
  parametric dispersion fit was wrong since I "fixed" the factor
  to get log2 asymptotic behaviour.

version 1.7.1 - 1.7.5:

- various changes to vignette by whuber

version 1.6.0 / 1.7.0: (Summary of important changes between Bioc release 2.8 to 2.9)

- changes to dispersion estimation scheme to improve handling of outliers
  (see vignette for details)

- refactorized the workflow and overhauled and expanded the vignette (see there
  for changes to user interface)

- various changes and bug fixes (see appendix in vignette and NEWS file)

version 1.5.29: (2011-10-10)

- fixed a bug in parametricDispersionFit. (Thanks to Alejandro Reyes for spotting and fixing this one.)

version 1.5.27: (2011-10-18)

- method="pooled" is now default in estimateDispersions
- added 'fitInfo' accessor function

version 1.5.24: (2011-08-10)

- fitNbinomGLM no longer stops when glm.fit throws an error

version 1.5.23: (2011-07-14)

- fixed bug in 'nbinomTest' again; now using a symmetric two-tailed test (i.e.,
  the one-tailed p value is doubled)
  
- added warning in case of method='blind'  if not used together with
  sharingMode='fit-only'

version 1.5.22: (2011-07-14)

- fixed bug in 'nbinomTest' that led to wrong p values for dispersions larger than 1

version 1.5.21: (2011-07-14)

- minor changes to vignette

version 1.5.20: (2011-07-07)

- fixed bug in 'getVarianceStabilizedData'

- more informative error message if parametric fit fails

versions 1.5.1 to 1.5.19: (2011-06 / 2011-07)

refactorized the way how dispersion estimation is handled, overhauled the vignette

- vignette now uses data from 'pasilla' package as example

- GLMs explained in vignette

- 'estimateVarianceFunctions' renamed to 'estimateDispersions'

- estimateDispersions now stores intermediate results in 'fitInfo' slot

- new functionality for 'estimateDispersion': 'sharingMode' option with new
  sharing modes 'maximum' and 'gene-est-only'; 'fitType' with new
  fitting type 'parametric'; changed defaults for these; new method
  'pooled-CR' to do edgeR-style CR estimation etc.
  
- several diagnostic functions removed

- lots of further small changes  


version 1.3.3: (2011-03-08)

- added 'locfunc' argument to 'estimateSizeFactor' and 'estimateSizeFactorForMatrix'

version 1.3.2: (2011-03-08)

- added 'glmControl' argument to 'nbinomFitGLM' and 'nbinomGLMsForMatrix'

version 1.3.2: (2011-02-11)

- updated CITATION

version 1.1.13: (2010-09-06)

- corrected the sometimes missing names attribute in rawVarFuncTable

version 1.1.12: (2010-09-06)

- fixed a bug in the 'resVarB' column of the result of 'nbinomTest'.

version 1.1.11: (2010-08-17)

- fixed a bug in getVarianceStabilizedData; it failed to preserve 
normalization. This is because the transformation function
for each sample were different if the size factors differed. The
function now uses an average size factor. However, the veriance 
stabilizing transformation is still not perfect:
the variance still increases slightly with the mean. 
This needs a bit more work

As an alternative to judge similarity of samples, I have added 
two new functions, getRawScvForSamplePair and 
getRawScvDistanceMatrix. See their help pages.

version 1.1.10: (2010-08-17)

- fixed bug in makeExampleCountDataSet: colnames were wrong

version 1.1.9: (2010-07-21)

- removed the variance adjustment factors introduced earlier; they caused
  more trouble than they were worth.

version 1.1.8: (2010-07-09)

- GLM functionality added but not yet documented

version 1.1.7: (2010-06-05)

- no change, just bumped version number as the build server seemed to have
  gotten confused

version 1.1.6: (2010-06-02)

- changed arguments for estimateVarianceFunctionsin preparation for a planned new
  feature; instead of pool=FALSE or pool=TRUE, use method="normal" or 
  method="blind".

version 1.1.5: (2010-06-01)

- fixed a small bug that caused p values to sometimes slighly exceed 1.0
  due to rounding errors

version 1.1.4: (2010-05-15)

- corrected a serious documentation bug: the effect of the variance
  adjustment factors was described the wrong way round

- changed scvPlot: now does a bias correction by default

version 1.1.3: (2010-05-06)

- added a CITATION file

version 1.1.2: (2010-05-03)

- replaced the C function to calculate the p values. The new version is now
  a bit more conservative with its approximation, trying to ensure that p values
  are always rather approximated as too high. Furthermore, it considers the
  values 0 and kS as critical points, too, which makes a difference at very 
  high overdispersion.
  
- changed the default value for the accuracy guidance 'eps' from 3e-3 to 1e-4, as
  the former seemed to have been a bit too high. (The approximation broke down
  in rare cases.)
  
- The 'eps' parameter is now exposed in 'nbinomTest' as well.    

version 1.1.1: (2010-04-29)

- added "variance adjustment factors" in order to allow for manual scaling
  of the raw variance estimate in case that the residual ECDF plots indicate
  a bad fit
  
- fixed a bug which caused the resVarB column of the data frame returned
  by nbinomTest to be wrong
  
- changed the scvPlot function to now show conditions rather than raw variance
  functions
  
- changed the way how 'rawVarFunc' interprets its second argument

- updated the vignette