PRTools contents

PRTools manual



Branch and bound feature selection

    W = featselo(A,CRIT,K,T,FID)

 A input dataset
 CRIT string name of the criterion or untrained mapping  (optional, def= 'NN' 1-Nearest Neighbor error)
 K numner of features to select (optional, def: K=2)
 T validation set (optional)
 N Number of cross-validations (optional)
 FID File ID to write progress to (default [], see PRPROGRESS)

 W output feature selection mapping


Backward selection of K features by baktracking using the branch  and bound procedure on the data set A. CRIT sets the criterion  used by the feature evaluation routine FEATEVAL. If the data set T is given, it is used as test set for FEATEVAL. Alternatively a number  of cross-validations N may be supplied. The resulting W can be used for  the selecting features of a dataset B by B*W.  The selected features are stored in W.DATA and can be found by +W.

This procedure finds the optimum feature set if a monotoneous  criterion is used. The use of a testset does not guarantee that.


P. M. Narendra and K. Fukunaga A Branch and Bound Algorithm for Feature Subset Selection, IEEE Trans. Computer, 26(9), pp. 917-922, September 1977

See also

mappings, datasets, feateval, featself, featselb, featseli,


PRTools contents

PRTools manual