perClass Documentation
version 5.4 (7-Dec-2018)
SDFEATSEL Feature selection


     PF=SDFEATSEL(DATA,DEF)   % Define feature subset manually

 Define feature subset by a trained decision tree PT
 Find features with zero variance
     PF=SDFEATSEL(DATA,'var>0')  % non-zero variance features

    DATA     Input data set
    DEF      Indices of features in DATA or cell array with names

    PF       Feature selection pipeline
    RES      Structure with detailed information on selection process

   'method'  Selection method (default: 'forward')
     'individual'  - Individual feature ranking
     'forward'     - Greedy forward search
     'backward'    - Greedy backward search
     'floating'    - Series of forward/backward searches
       'rounds'    -  Number of floating rounds (default: 10)
     'random'      - Best solution from a set of randomly generated subsets
       'count'     -  Number of random solutions (default: 200)
       'bounds'    -  Vector [min,max] number of features taken randomly
   'model',M   Use error of untrained pipeline M as criterion
   'from'      Initial solution for forward, backward or floating search
   'steps'     Make N steps and return best subset (only forward and backward)
   'test'      External test set used for criteria evaluation
   'trfrac'    Fraction of DATA used for training (default: 0.75)
   'nodisplay' Do not show any output

 SDFEATSEL selects a subset of features of the the data set DATA. By
 default SDFEATSEL minimizes the error of 1-NN classifier. Any untrained
 classifier may be supplied in 'model' option.  By default, the forward
 greedy search is performed. The classifier is trained on 75% and tested
 on the rest of DATA.
 User may limit the number of steps performed by forward and backward
 searches with 'steps' option. Best solution is always returned.
 Floating search combines several rounds of full forward and backward
 search. By default it is initialized from a random search (use 'from'
 option to specify initial subset manually). Subsets found in floating
 search are returned in RES.feat cell array.


sdfeatsel is referenced in examples: