=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.3.2 ---------------------------------

Mon 27 Sep 2010 10:48:55 AM CEST
====================================================

Mon 27 Sep 2010 09:56:24 AM CEST
train_sc:
    fix SVM for data with missing values (deletion mode) 

Tue 21 Sep 2010 09:57:33 PM CEST
xptopen:
    SPSS: decode some more header information - still not ready

Sun 19 Sep 2010 11:34:13 PM CEST
train_sc:
    remove obsolete test for weight vector in LibLinear


=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.3.1 ---------------------------------

Wed 08 Sep 2010 12:17:18 PM CEST
====================================================

Sun 05 Sep 2010 01:34:22 AM CEST
xptopen:
    reading ARFF files supported;
    SPSS: better identification

Mon 30 Aug 2010 10:25:15 PM CEST
xptopen:
    support for reading STATA files added

Mon 30 Aug 2010 11:15:15 AM CEST
xptopen:
    remove obsolete code; 
    improve reading variable names (prevents seg-fault in some special case); 
    write: put XPTOPEN into header; 
    experimental date/time conversion


=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.3.0 ---------------------------------

Fri 27 Aug 2010 12:57:22 PM CEST
====================================================

Thu 26 Aug 2010 11:27:16 AM CEST
test_sc: 
    improve docu: change 'distance' to 'signed distance' 
        and explain meaning of a negative distance

Tue 24 Aug 2010 11:14:40 PM CEST
xptopen: <NEW>
    reading and writing of SAP Transport file format *.xpt
ttest, ttest2: <NEW>
    support of paired and unpaired t-test for data with missing values
    make it compatible to TMW's interface

Thu 12 Aug 2010 11:03:10 PM CEST
train_sc:
    fix for single column classlabel with elements {-1,1}

Thu 12 Aug 2010 09:25:42 PM CEST
doc/INSTALL
    add more detailed description on installing mex-files

=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.2.1 ---------------------------------

Thu 29 Jul 2010 10:01:36 PM CEST
====================================================

Wed 28 Jul 2010 11:29:34 AM CEST
median, kth_element, naninsttest:
    addresses undesired side-effect off in-place sorting of data

Fri 25 Jun 2010 11:49:51 PM CEST
kth_element:
    improve speed, avoid recursion, simplify code

=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.2.0 ---------------------------------

Fri 25 Jun 2010 12:11:46 PM CEST
====================================================

Fri 25 Jun 2010 11:12:14 PM CEST
docu improved
quantile:
    empty set returns NaN
xcovf:
    improve compatibility
mexw32
    binaries included

Thu 24 Jun 2010 09:11:48 PM CEST
median: 
    fix typo

Tue 08 Jun 2010 10:59:09 PM CEST
kth_element <NEW>, median:
    add kth_element, this improves performance of median

Tue 01 Jun 2010 09:00:43 PM CEST
row_col_deletion:
    row_column_deletion algorithm to exclude missing values is now more flexible
    rows and columns can be selected simultaneously - overcome the fact that 
    only either rows or columns can be selected.

Fri 14 May 2010 03:13:26 PM CEST
histo_mex:
    avoid using ssize_t
    add support for char-data 

Mon 26 Apr 2010 04:16:03 PM CEST
str2array: 
    fix uninitialized u[k] causing seg-faults
make.m
    fix types for octave

Sat 24 Apr 2010 02:21:01 AM CEST
str2array:
    str2double renamed to str2array
    extra space characters are ignared
liblinear/train.mex
    fix making for win32/matlab platform
rem,mod:
    workaround for bug in Octave's repmat

Tue 30 Mar 2010 08:48:20 PM CEST
str2double:
    fix default row delimiter 

Thu 25 Mar 2010 02:10:59 PM CET
make: 
    fix for str2double
src/liblinear
    fix for matlab v71

Tue 23 Mar 2010 08:08:45 PM CET
xcovf: 
    fix for SCALEOPT='coeff'
corrcoef:
    reverse reshaping of nan_sig incase of non-empty Y; 
    set isnan(R) to zero;

Sat 13 Mar 2010 08:52:49 AM CET
corrcoef:
    fix output argument nansig for non-empty Y

=== Package: NaN-toobox for Octave and Matlab ======
--- Version: 2.1.0 ---------------------------------

Sun 28 Feb 2010 09:52:51 PM CET
====================================================

Wed 24 Feb 2010 11:10:33 PM CET
str2double: 
    support for cell arrays and complex numbers added

Sun 21 Feb 2010 10:25:16 PM CET
str2double: 
    mex implementation of str2double added for performance reasons

Wed 27 Jan 2010 08:43:30 PM CET
train_sc:
    several LDA improvements: ill-defined covariances better supported, a litte speed-up
covm:
    remove warning because its not needed.

Mon 11 Jan 2010 09:48:07 PM CET
liblinear:
    supported added:
    weighted liblinear supported

Sun 10 Jan 2010 10:22:42 PM CET
libsvm: 
    add sources for libSVM - this enables native SVM-support without additional dependencies
    sources need to be compiled with g++ 

Sun 10 Jan 2010 01:13:34 AM CET
train_sc, test_sc, xval, test_xval: 
    {-1,+1} encoding of classlabel supported; 
    test of liblinear and libSVM with Octave3.2

Fri 08 Jan 2010 02:09:21 PM CET
*.m: 
    major cleanup of syntactic issues reported by mlint. 

Fri 08 Jan 2010 12:05:56 PM CET
train_sc: 
    fix nested functions for matlab 
test_fss: 
    report improved

Thu 07 Jan 2010 11:19:58 PM CET
test/test*.m
    add several functions for testing of the NaN-toolbox 

Thu 07 Jan 2010 09:53:04 PM CET
train_sc: 
test_sc:
    add Adaline/LMS and fix PLA 
    add ###/DELETION modes - this supports missing values 
	also in otherwise unsupported classifiers like 
	PLA, SVM, LMS, sparse, GSVD

xval: used standard kappa (ignore NaNs)

Thu 07 Jan 2010 02:15:10 PM CET
train_sc: 
test_sc:
    fix for PLA 
    hyperparameters renamed to hyperparamter (both were used before)
    
Thu 07 Jan 2010 09:12:28 AM CET
train_sc: 
test_sc:
    disable PLA
    fix FLDA

several minor changes 

Mon 07 Dec 2009 11:51:50 PM CET
fss:
    literature check and improve docu - it seems the proposed method is new

Sun 29 Nov 2009 08:16:03 PM CET
partcorrcoef:
    use as default a more robust estimation method

Fri 27 Nov 2009 03:35:46 PM CET
fss:
    partcorrcoef replaced by a more robust version

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.0.5 ----------------------------------  

Mon 23 Nov 2009 04:31:34 PM CET
====================================================

Mon 23 Nov 2009 11:57:53 AM CET
nanmean: 
    fixed
tinv: 
    check of input arguments fixed 

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.0.4 ----------------------------------  

Sat 21 Nov 2009 11:53:44 PM CET
====================================================

Sun 08 Nov 2009 09:26:04 PM CET
xval:
    fix simple case xval(D,C)
trimmean: 
    fix usage

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.0.3a ----------------------------------  
fixed: std, var; 
removed: mod, rem

Sat 07 Nov 2009 09:23:00 PM CET
====================================================

Sat 07 Nov 2009 08:32:29 PM CET
rem,mod: 
    keep it in the repository, but do not include it in the releases
    
Thu 05 Nov 2009 04:03:14 PM CET
rem,mod: 
    removed in order to remove warning messages that confuse the user

Mon 26 Oct 2009 09:13:31 PM CET
rem,mod:
    check of input arguments improved

Fri 16 Oct 2009 12:49:19 AM CEST
std,var:
    center ignoring weighted samples fixed
    speed-up 2nd method
    
Wed 14 Oct 2009 12:45:03 AM CEST
sumskipnan_mex, histo_mex, covm_mex: 
    fix version check for defining mwSize; 
    set default ACC_LEVEL 0

Tue 06 Oct 2009 12:57:11 PM CEST
var:
    fix for opt=[]

Thu 01 Oct 2009 02:14:10 PM CEST
acctest: <NEW> 
    test to determine optimal accuracy level
flag_accuracy_level: 
    default accuracy level set to 0;
    improve docu 
sumskipnan_mex, covm_mex    
    level 2 and 3 exchanged; 
    remove debugging information
    improve docu

Wed 30 Sep 2009 10:18:37 AM CEST
sumskipnan_mex:
    increase accuracy by using Kahan's summation formula. 
    This reduces the error by a factor of N; first tests show a penalty of about 40% in terms of speed; 
    Currently, this is only implemented for stride=1 (summation along 1st dimension)

Tue 29 Sep 2009 05:29:28 PM CEST
covm_mex:
    increase accuracy by using Kahan's summation formula 
    this reduces the error by a factor of N; 
    at a cost of a 25% slowdown. 
    

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.03 ----------------------------------  

Thu 24 Sep 2009 10:43:37 AM CEST
====================================================

Tue 22 Sep 2009 06:14:01 PM CEST
xval: 
	fix ambiguity between W and NG if W is empty; 
        classlabels can be cell array, too.
load_fisheriris: <new>
	support loading of famous Fisher(1936) iris data set

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.02 ----------------------------------  

Fri 04 Sep 2009 05:01:22 PM CEST
====================================================

Wed 02 Sep 2009 03:25:12 PM CEST
partcorrcoef:
    remove obsolete warning
cat2bin: 
    fix handling NaN's in default mode
    add support for modes 'IgnoreZeros' and 'IgnoreZeros+NaN' 

Sun 30 Aug 2009 12:03:27 AM CEST
train_sc
    PSVM: experimental weighting of data samples

Tue 11 Aug 2009 09:54:43 AM CEST
histo_mex
    support for logical data added
    
Thu 06 Aug 2009 03:45:10 PM CEST
histo_mex
    fix H.X(256) for uint8 data
    
Wed 05 Aug 2009 09:48:02 PM CEST
train_sc:
    perceptron learning algorithm (PLA) and Winnow algorithm added 
    
Wed 05 Aug 2009 03:08:03 AM CEST
fss: <NEW>
    feature subset selection and ranking added

partcorrcoef: 
    fix case if Z is empty

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.01 ----------------------------------  

Tue 04 Aug 2009 11:48:40 AM CEST
====================================================

Tue 04 Aug 2009 11:48:18 AM CEST
train_sc:
    bug fix 

Thu 30 Jul 2009 08:09:20 AM CEST
covm_mex: 
    prevent seg-fault for the case of X and Y have same starting location, but different size
covm: 
    remove redundant code 
    fix computation of mean of complex data
    

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 2.00 ----------------------------------  

Wed 29 Jul 2009 08:10:44 AM CEST
====================================================

Mon 27 Jul 2009 10:29:48 PM CEST
sumskipnan_mex:
    avoid seg-fault if empty input argument

Mon 27 Jul 2009 02:42:06 PM CEST
corrcoef
    fix warning status 

Thu 02 Jul 2009 02:22:52 PM CEST
    add classification methods, 
    change license to GPL v3 or later 
    version number 2.0

Tue 30 Jun 2009 02:12:10 PM CEST
sumskipnan_mex
covm_mex
    using internally long double for increased accuracy

Mon 15 Jun 2009 10:22:57 AM CEST
histo_mex
    support for data weightening added 

Wed 10 Jun 2009 02:19:43 PM CEST
    faster exist() improves performance (especially in covm.m)

Mon 08 Jun 2009 01:12:16 PM CEST
histo_mex
    computation of histogram added

Wed 03 Jun 2009 10:23:30 PM CEST
covm_mex
    minor improvements
    
Wed 27 May 2009 11:50:20 PM CEST
sumskipnan_mex
covm_mex
    support for Matlab v7.2 and earlier 

Mon 25 May 2009 03:44:27 PM CEST
corrcoef
    extended input arguments 

Fri 22 May 2009 02:52:32 PM CEST
flag_implicite_skip_nan
flag_implicite_significance_level:
    replace global with persistent variable 

Wed 20 May 2009 01:33:09 AM CEST
IQR, TRIMMEAN, PRCTILE
    added
QUANTILE
    allow for 4th input argument    

TWed 13 May 2009 04:36:13 PM CEST
ECDF, TIEDRANK, PARTCORRCOEF, zScoreMedian  medAbsDev
    added 

ue 12 May 2009 04:03:42 PM CEST
rms, sem, std, var
    support weighting of data 
flag_implicit_skip_nan
    tested and hacked for the use with Freemat v3.6

Mon 11 May 2009 10:42:29 AM CEST
covm:
    bug fix normalization of Mode='E' in case mexFLAG

Mon 11 May 2009 09:02:26 AM CEST
center, mean, geomean, harmmean, meansq, zscore 
    support weighting of data 

##### Release of Octave-forge from Soren Hauberg 

Thu 07 May 2009 09:30:00 PM CEST
covm_mex: 
    faster if X==Y, exploiting the fact that the output is symetric; 
    docu improved

Thu 07 May 2009 11:28:46 AM CEST
covm_mex: 
    performance test of alternative algorithm, and optimization for speed; 
covm.m
    support complex matrices; 
sumskipnan_mex
    fix SSQ if weighted data

Wed 06 May 2009 06:16:33 PM CEST
covm_mex
    mex version for covm added 

Wed 06 May 2009 01:04:11 PM CEST
sumskipnan_mex
sumskipnan
covm
    support weighting of samples 

ranks
    remove clumsy code for support of 2.0.16
    this is needed to work with freemat

Tue 05 May 2009 01:51:49 PM CEST
sumskipnan_mex
    correct type of size 

Wed 15 Apr 2009 09:47:10 AM CEST
tcdf
    bug fix for x=inf

Tue 17 Mar 2009 09:24:24 AM CET
flag_implicit_skip_nan
    added (again). The function can be used to turn off and on the 
    NaN-skipping behaviour. Turning off skipping NaN's 
    with flag_implicit_skip_nan(0) gets also rid of the 
    performance penalty for testing NaN's. Thus, the 
    traditional behavior can be restored. 
    

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 1.91 ----------------------------------  

Wed 11 Mar 2009 09:18:30 AM CET
====================================================

Wed 11 Mar 2009 09:01:28 AM CET
SUMSKIPNAN_MEX: 
    minimize cache misses - increases performance; 
    support for complex data moved to sumskipnan.m

=== Package: NaN-toobox for Octave and Matlab ======  
--- Version: 1.90 ----------------------------------  

Tue 10 Mar 2009 01:51:23 PM CET
====================================================

Tue 10 Mar 2009 01:30:03 PM CET
    flag_nans_occured is not reset at every sumskipnan but only at the 
    next call to sumskipnan; this enables a flexible control of the 
    granularity on explicit NaN-checks
    
Mon 09 Mar 2009 12:03:43 PM CET
SUMSKIPNAN_MEX:
    much faster
    support of flag_isnan
    
Fri 06 Mar 2009 12:58:21 PM CET
FLAG_NAN_OCCURED, SUMSKIPNAN, COVM:
    flag_nan_occured added to indicate whether some NaN's 
    appeared in the last input data 

Thu 05 Mar 2009 12:57:40 PM CET
    sumskipnan.oct was incomplete causing some problems. 
    In order to solve this issue, 
    sumskipnan.cc renamed to sumskipnan_oct.cc
    sumskipnan.cpp renamed to sumskipnan_mex.cc        

Wed 17 Sep 2008 08:49:24 AM CEST
NANSTD:
    evaluate with default method if flag is empty

Tue 28 Aug 2007 10:17:07 AM CEST
QUANTILE, PERCENTILE: 
    bug fix for quantile([1,2,4,700],[0.05,0.99])
    arrays and histograms: q<=0,q>=1 addressed, 
    mean of upper and lower bound is returned if case of non-unique solution

Fri 06 Oct 2006 09:57:52 AM CEST
ZSCORE
    + pre-whitening support, scaling factors are returned
    + use two-pass algorithm to increase accuracy
CORRCOEF
    + indexing improved: if col(X)*col(Y) << max(col(X),col(Y))^2

Sun 06 Aug 2006 10:31:11 PM CEST
NANSTD
    define FLAG if nargin=2 
STD, VAR
    replace algorithm by more accurate but slower 
    two-pass algorithm

Wed 02 Aug 2006 07:21:03 PM CEST
NANSTD
    add FLAG argument for compatibility with Matlab 

Wed 05 Apr 2006 08:37:54 AM CEST
STD
    force positive result. 

Thu 16 Mar 2006 10:38:11 CET
QUANTILE.M
PERCENTILE.M
MEDIAN
    minor change: more efficient index use

Tue 07 Mar 2006 00:46:48 CET
QUANTILE.M
PERCENTILE.M
    support of DIM arguments (N-dimensional matrices)

Mon 03 Oct 2005 19:18:18 CEST
CENTER.M
    docu improved 

Thu 02 Jun 2005 14:23:40 CEST
SUMSKIPNAN.M
    replace SUM by X*ONES for performance reasons
    replace ISNAN(X) by X~=X for performance reasons

COVM 
    replace ISNAN(X) by X~=X for performance reasons


Wed 25 May 2005 18:03:59 CEST
Copyright cleanup by Paul Kienzle 

Wed 11 May 2005 17:34:57 CEST
PERCENTILE, QUANTILE
    replace INTERP1 by FLIX because
    - interp1 does not work for single element Y
    - requires 3 input arguments in Octave 

Fri 29 Apr 2005 00:03:13 CEST
SUMSKIPNAN.CC
    support for NDArrays added

Fri 22 Apr 2005 23:31:03 CEST
SUMSKIPNAN.CC
    OCT-version added 
    First tests show a speedup by a factor between 4 and 5 

Tue 19 Apr 2005 15:23:48 CEST
XCOVF
    added

============================================
Wed 16 Mar 2005 15:22:03 CET

Release 1.60
===========================================

Wed 16 Mar 2005 15:21:27 CET
PERCENTILE, QUANTILE
    replace FLIX by INTERP1

PERCENTILE
    remove FLAG_DEFAULT_SKIP_NAN

Wed 16 Mar 2005 00:46:02 CET
RANKS
    conversion of version number fixed for 2.9.0

============================================
Thu 03 Mar 2005 10:10:48 CET

Release 1.59
===========================================

Thu 03 Mar 2005 10:10:17 CET
SUMSKIPNAN.CPP 
    fixed compilation error     

Wed 02 Mar 2005 19:32:31 CET
COV	
    address incompatibility to traditional matlab implementation. 

============================================
Wed 23 Feb 2005 03:47:41 PM CET

Release 1.58
===========================================

Sat 15 Jan 2005 04:20:31 AM CET
NANTEST
    test for ALL and ANY added

Tue 07 Dec 2004 12:54:27 AM CET
NANTEST
    Test (0-3i)/inf included

Wed 27 Oct 2004 09:11:43 CEST
COR
    fix for ML7

Mon 05 Jul 2004 12:39:41 PM CEST
MEDIAN
    type in docu corrected 

============================================
Thu 10 Jun 2004 08:52:48 PM CEST

Release 1.55
===========================================

Thu 10 Jun 2004 08:46:23 PM CEST
CORRCOEF
    nargout is not reentrant in Octave. 
    This caused [r,p]=corrcoef(x,'rank') did 
    not return a p-value. FIXED. 

Wed 11 Feb 2004 08:30:18 PM GMT+1
MOD.M, REM.M
    performance improvments
    - sign test not needed
    - error interval only calculated when needed. 

Fri 30 Jan 2004 02:53:05 PM CET
COR, CORRCOEF
    improve ducumentation. 
    abs(COR) can be larger than 1.
    abs(CORRCOEF) is never larger than 1. 
        
Fri 30 Jan 2004 03:25:55 AM CET
MOD.M, REM.M
    alternative definition of MOD and REM
    The most important difference are that 
        MOD(X,0)=0; MOD(X,Inf)=X;
	REM(X,0)=0; REM(X,Inf)=X; 
    Furthermore, an error tolerance is returned, too. 
        [Z,E]=MOD(X,Y); with Z+-E indicating 
	the interval and accuracy of Z	
	E > Y indicates an intolerable accuracy. 
 	
Thu 29 Jan 2004 08:47:12 PM CET
NANTEST
    check whether BITAND and BITSHIFT can return NaN.
    check REM and MOD for their behavior with extrem values

Tue 23 Dec 2003 02:10:52 PM CET
CONV2NAN.m added
    2-dimensional convolution of data containing NaN's
    Naming it CONV2 would not work because CONV2 is built-in. 
    
============================================
Thu 04 Dec 2003 01:54:51 PM CET

Release 1.54
===========================================

Thu 04 Dec 2003 01:54:51 PM CET
SUMSKIPNAN.DLL
    removed because ML6.5 compilate does not work on ML5.3

Thu 27 Nov 2003 12:16:36 PM CET
QUANTILE
    FLAG_IMPLICIT_SKIP_NAN removed
 
Wed 26 Nov 2003 01:55:35 AM CET
Successful test of NDarrays in Octave 2.1.52 

Tue 25 Nov 2003 10:11:17 PM CET
SEM
    prevent round off error causing imaginary resutls 

Tue 25 Nov 2003 02:45:28 AM CET
NANTEST
    test of support for ND-arrays included

Thu 20 Nov 2003 20:25:31 PM CET
Test of NDarrays on Octave 2.1.51 failed for the following functions
    ISNAN, ISINF, FINITE, SORT, DIFF
    submitted a bug report.

Wed 05 Nov 2003 01:51:48 PM CET
FLAG_IMPLICIT_SKIP_NAN 
    removed

============================================
Tue 04 Nov 2003 08:38:20 PM CET

Release 1.53
===========================================
Wed 05 Nov 2003 01:46:31 PM CET
SEM.M	
    fix second output argument

Fri 31 Oct 2003 09:33:54 PM CET
SUMSKIPNAN2.MEX renamed to SUMSKIPNAN.MEX
    accordingly, the MEX-file instead of M-file is used.     

SUMSKIPNAN: 
    Profiling showed that EXIST and FLAG_IMPLICIT_SKIP_NAN
    are significant factors for a slowdown of SUMSKIPNAN. 
    Therefore, SUMSKIPNAN.M is simplified and the functions
    EXIST and FLAG_IMPLICIT_SKIP_NAN are removed. 

Mon 27 Oct 2003 08:34:47 PM CET
MEAN, VAR
    docu corrected
    
Thu 09 Oct 2003 12:50:39 PM CEST
ZSCORE
    fix bug for complex data. 

============================================
Mon 06 Oct 2003 03:57:20 PM CEST

Release 1.52
===========================================
included pre-compiled mex-files of SUMSKIPNAN2

Mon 06 Oct 2003 02:52:52 PM CEST
TRIMEAN
    replace flix

Mon 29 Sep 2003 11:32:16 AM CEST
ZSCORE
    increase efficiency of complex data

Sat 27 Sep 2003 01:51:40 PM CEST
SUMSKIPNAN2.CPP
    fix possible memory leak 
    instability in LNX86 Matlab 5.3 (R11) still observed.

RMS, SEM, NANSTD, MEANSQ
    increase efficiency of complex data
    
============================================
Fri 26 Sep 2003 02:49:14 PM CEST

Release 1.51 	
============================================

Fri 26 Sep 2003 11:33:48 AM CEST
SUMSKIPNAN2.CPP
    summed square also for complex data 
SUM,VAR,STD,CENTER
    check of input arguments simplified
SUM,VAR,STD
    Use of sum square of complex data improved   

Thu 25 Sep 2003 06:43:13 PM CEST
SUMSKIPNAN2.CPP
    check DIM argument improved
STD
    implements best unbiased estimation of the Standard Deviation
    NOTE: this is different than the square root of the 
	    best unbiased estimate of the variance

Wed 24 Sep 2003 06:25:47 PM CEST
NANTEST
    test for GEOMEAN and HARMMEAN included
    
Mon 22 Sep 2003 10:03:53 PM CEST
NANTEST & NANINSTTEST
    Tests for buggy NANSUM and NANSTD (from Mathworks) included
NANSUM, NANSTD
    Fixes included

Mon 22 Sep 2003 08:09:11 PM CEST
SUMSKIPNAN2.CPP	
    support for real ND-arrays included
    support for complex arrays included

Thu 18 Sep 2003 07:16:48 PM CEST
MEDIAN, TRIMEAN
    support of ND-arrays implemented    

Thu 18 Sep 2003 06:19:51 PM CEST
SUMSKIPNAN2.CPP	
    included source of C-MEX file 

Thu 18 Sep 2003 12:35:12 PM CEST
SUMSKIPNAN.M
    support of Version < 2.1.36 discontinued
    support of MEX-implementation included 

Mon 15 Sep 2003 06:08:37 PM CEST
NANINSTTEST	
    missed some bugs, NANTEST found already - CORRECTED

NANTEST 
    test for strange behaviour of Matlab included
      (division of two singluare matrices returns an Inf-Matrix in Matlab)
COVM	
    use of tempary variable makes code more memory efficient 
    - larger matrices can be handled

    
============================================
Sun 14 Sep 2003 04:28:54 AM GMT+1

Release 1.50 	
============================================


Fri 12 Sep 2003 04:13:30 PM CEST
TPDF, TCDF and TINV: 
	included 
    	these are needed to fix some Matlab bux
NANTEST: 
	test of TPDF, TCDF and TINV included; 
	test of Matrix Division Error NaN/NaN modified for ML6.5 - Dimension must be p=3 or larger
NANINSTTEST: 
	test for NORMPDF, NORMCDF, NORMINV, TPDF, TCDF and TINV included

Thu 11 Sep 2003 06:08:40 PM CEST
TCDF included; 
    fixes a problem of TCDF from Stats-TB 4.0. 
    This has caused problems with the significance test 
    of Spearmans correlation. 
    
Wed 10 Sep 2003 04:13:30 PM CEST
NANTEST
    NANTEST separated; 
	NANTEST checks only NAN-related properties of standard Octave;
	NANINSTTEST checks whether the NaN-tb is correctly installed. 
	
Thu 17 Jul 2003 09:04:44 PM CEST
CENTER	
    2nd output argument returns mean
NANTEST
    warning flag saved and turned off
        

Mon 30 Jun 2003 06:34:23 PM CEST
CORRCOEF:
    test with Octave 2.1.42, ML5.3, ML6.5
	    

Son 29 Jun 2003 23:46:18 AM CEST
CORRCOEF: 
    significance test for checking if NaN's are correlated 


Tue 27 May 2003 05:47:53 PM CEST
VAR, STD:
    arg2 (OPTION) included, DIM is now third argument. 
    This makes it compatible (but not equal) to Matlab versions. 


============================================
Sat 10 May 2003 09:08:58 PM CEST

Release 1.40 	

NaN-tb is now available from http://sf.net/projects/biosig/ 
together with TSA 
============================================


Tue 15 Apr 2003 12:09:25 PM CEST
NANTEST include test on sorting NaN's

PERCENTILE added

FLAG_IMPLICIT_SKIP_NAN included again. It is useful for temporary turn-off of SKIP-NAN
it is used in:  SUMKSKIPNAN, PERCENTILE, QUANTILE, MEDIAN, TRIMEAN



Thu 13 Mar 2003 07:05:56 PM CET
COVM, NORMPDF, NORMINV, NORMCDF, COEFFICIENT_OF_VARIATION
       Change 1st comment line, HELP NAN looks better now. 



Sun 02 Mar 2003 11:53:53 PM CET
COV  
	restore compatibility to Octave



Fri 28 Feb 2003
RANKCORR 
	spelling in documentation corrected, revision and id included
STD, VAR
	documentation improved		



============================================
Tue 25 Feb 2003 03:41:57 PM CET

Release 1.30 	
============================================


CHANGELOG started
previous changes can be obtained from octave-forge