NEWS for 'nan' Package

Return to the 'nan' package

2022-08-16: Release of NaN-toolbox 3.7.1

* kurtosis, skewness:
    add support for FLAG input argument, supporting
    biased and unbiased estimation.
    Compatibility tests with Octave and Matlab are added.
    It will break backwards compatibility, the previous
    version is supported by FLAG=-1.

* fishers_exact_test:
    extend to non-square contingency tables according to
    https://mathworld.wolfram.com/FishersExactTest.html


2022-05-09: Release of NaN-toolbox 3.7.0

* minor improvements (kappa, cdfplot, signrank)
* fix installation when matlab is in non-default location


2021-11-03: Release of NaN-toolbox 3.6.1

* signrank: update computation of p-value
    according to revised references,
    and validated against R.
* cdfplot: use histo instead of histo_mex


2021-07-26: Release of NaN-toolbox 3.6.0

* corrplot, knnsearch <NEW>


2021-02-15: Release of NaN-toolbox 3.5.3

* kappa: Youden index (YI), and SSE as return value
* roc: add maxYI as output value
* naninsttest: check ttest for NaN handling


2020-10-31: Release of NaN-toolbox 3.5.2

* new functions added
   gini: Gini index
   lmom: L-moments
   betapdf, betacdf, betainv:
	beta distribution functions

* bug fixes:
  bland-altman: fix ylabel
  fix debian bug #964210: test tries to download data
  fix #59335: "'load_fisheriris' fails on Windows"
  fix #59336: "(nan) BIST for `ttest` fails"
    this failed because betainv(...) is not available
    the suggestion of add a dependency on statistics package
    does not work, because of a conflict with statistics
    (it would not handle NaNs in the indended way).
    Therefore, functions for beta* distributions are included.


2020-07-12: Release of NaN-toolbox 3.5.0

* ttest:
  - support output arguments CI and STATS

* fishers_exact_test:
  - use VPA from symbolic pkg if available; 
  - add unit tests; 
  - improve help   


2019-11-16:  Release of NaN-Tb v3.4.5

- fishers_exact_test: added <NEW> 

- bug fixes for:
  #57263: [octave forge] (nan) fails to build on a system without libblas (Octave built with OpenBLAS)
  #57232: [octave forge] (nan) Spelling error in message of function str2array
  #50248: [octave forge] (nan) Spelling error in the help string of the train function
  #57228 "[octave forge] (nan) fails to build with undefined references in Octave 5 when LDFLAGS is set"
- gscatter: fix use of argument "doleg"
- address "warning: no return statement in function returning non-void"

2019-10-27:  Release of NaN-Tb v3.4.3

- Makefile.*
    Add missing copyright/licenses

- liblinear, libsvm:
    add license from original author

- roc.m: number of data points limited to about 5000
    add final data point at (1,1).


2019-09-23:  Release of NaN-Tb v3.4

- quadratic classifiers:
  check for positive-definiteness of covariance matrix
  and return no classification (encoded as NaN), if not.

- mad, meandev:
  make sure source is ascii/utf-8 compliant


2019-06-08:  Release of NaN-Tb v3.3

- detrend: the 2nd output (trend) can be interpolated and should 
  not contain NaNs.

- train_sc: some supported classifiers should not be used 
     with the short-cut for two-class problems. 

- xval: report also results w/o crossvaliation

- ecdf: fix ylim in case data contains missing values.


2019-04-15:  Release of NaN-Tb v3.2.3

- <NEW> ROC is included. ROC is derived
  from the implementation in "biosig for octave and matlab"
  	biosig-code/biosig4matlab/t490_EvaluationCriteria/
  https://sourceforge.net/p/biosig/code/ci/master/tree/biosig4matlab/t490_EvaluationCriteria/

- <NEW> load_cifar100 load_cifar10 load_mnist:
  loading of various machine-learning databases

- <NEW> kolmogorov_smirnov:
   Multiple two-sample Kolmogorov-Smirnov test

- <NEW> kstest2:
   wrapper for kolmogorov_smirnov

- requires Octave 3.8 or later

- histo2, histo3: 
  bug fix for case when whole column contains NaNs

- <NEW> inst/signrank.m: add Wilcoxon signred rank test 
  Unlike Octave's wilcoxon_test, this works also for 
  sample sizes smaller than N=25

- corrcoef: improve documentation on one- 
  and two- pass algorithm

- histogram functions (histo.m) included

- upgrade to liblinear 2.2.1
  weightening of samples not supported anymore

- upgrade to libsvm 3.2.3
  weightening of samples not supported anymore

- minor changes


2017-08-23: Release of NaN-Tb v3.1.4

- minor fixes:
	license text, file permissions, 
	remove dependency on -lcholmod


2017-08-14: Release of NaN-Tb v3.1.3

- fix bug #43931: bug in help of corrcoef.m
	reported by: locker@vp...
- fix bug #47673: build failure for ppc64 architecture
	reported by: Michel Normand
- bug #48771 "Link against the shared library libsvm"
    had been fixed previously in Nov 2017

- tiedrank: increase matlab compatibility by allowing
    third input argument


2017-01-10: Release of NaN-tb v3.1.2

- fix build when Octave does not support openmp

2017-01-03: Release of NaN-tb v3.1.1

- add Makefile to built release

2016-11-30: Release of NaN-tb v3.1.0

- fix bug 48426: built for Octave 4.2.0
- prepare for using external libsvm and liblinear
- fix typos, address compiler warnings and improve docu

2016-08-10: Release of NaN-tb v3.0.3
- tcdf: fix bug #48731 (thanks to Nir Krakauer)
- coercoef: add comment on Matlab compatibiliy in help
- test_fss, load_fisheriris: 
          fix support on mingw platform (Windows)

2016-07-30: Release of NaN-tb v3.0.2
- Octave/Windows: setenv CC=gcc (fixes bug 47559)
- no OpenMP for Octave/Windows and MacOSX


2015-02-26: Release of NaN-tb v3.0.1

- fixes version number 
  two digit version number like 3.0 cannot be used in Octave-forge

2015-09-12: Release of NaN-tb v2.8.3

- ready for gcc/g++ v5 
- honor environment variables CC and CXX
- fix installer issue on OpenSuse
- support debian hardening flags
- internal improvents (in type and macro usage)
- address a number of compiler warnings
- bug fixes 

2015-07-06: Release of NaN-tb v2.8.1

- fix nantest to avoid crashing octave 4.0.0 on windows

2015-06-24: Release of NaN-tb v2.8.0

- fix Makefile for Octave4 on windows
  (fixes #45363 and #44859)

- check for sparse input matrices and
  convert to full when needed

2015-06-01: Release of NaN v.2.7.6

- improve accuracy of normcdf (bug #38170) 

2015-04-23: Release of NaN v2.7.5

- fix compiler issue with __isnan
- do not display diagnostic messages

2015-04-12

- upgrade to libsvm-3.12
- fix multi-threaded build (make -j) 
- improve some tests


2015-03-31: Release of NaN 2.7.2

- bug fix in nanstd.m 
- cross-compiler issues in histo_mex
- fix Makefile for use with Matlab
- address compatibility issue on debian/jessie
- minor issues (some compiler warnings are addressed)


2015-01-24: Release of NaN 2.7.1

- support for MacOSX added
  (tested on Homebrew Octave)


2015-01-17: Release of NaN 2.7.0

Fix compatibility issues with core functions in Matlab and Octave
- zscore: order of input output argument
- mahal gives now same result than matlab
- sumskipnan allows DIM argument larger than ndims
- a few minor issues

For changes of previous releases see: 
  http://pub.ist.ac.at/~schloegl/matlab/NaN/CHANGELOG

Package: nan