The NaN-toolbox

A statistics and machine learning toolbox for data with and w/o MISSING VALUES encoded as NaN's.

FEATURES of the NaN-toolbox:

Statistical tools
classification methods (including FDA/LDA, MDA/QDA, NBC/aNBC, REG/PLS, WienerHopf, PLA, SVM ...)
NaN`s are treated as missing values
supports weighting of data samples
yields more often the correct result (instead of NaN)
less but more powerful functions (no nan-FUN needed)
fixes known bugs of traditional functions
compatible to Matlab and Octave
significance test, confidence intervals and Spearman`s rank correlation included in CORRCOEF
CORRCOEF checks whether NaN's (missing values) are correlated with data
flexible control of granularity for explicit checks of NaN's
parallel execution on multicore platforms [1]
NEW: signrank: Wilcoxon's signed rank test, works also for N less than 25
NEW: histo2,histo3,histo4,roc,kappa (from TSA toolbox)
NEW: load_cifar100 load_cifar10 load_mnist