%FEATSEL Fixed mapping for selecting given features
%
% [W,V] = FEATSEL(K,J)
% B = A*FEATSEL(J)
% C = A*FEATSEL(J,true)
%
% INPUT
% K Input dimensionality
% J Index vector of features to be selected
% A Dataset
%
% OUTPUT
% W Mapping performing the feature selection
% V Mapping selecting the complementing features
% B Dataset, A(J,:)
% C Dataset, A with features J removed
%
% DESCRIPTION
% This is a simple support routine that writes feature selection
% in terms of a mapping. If A is a K-dimensional dataset and J are
% the feature indices to be selected, then B = A*W does the same as
% B = A(:,J).
%
% The use of this routine is a mapping V computed for a lower dimensional
% subspace defined by J can now be defined by W = FEATSEL(K,J)*V as a
% mapping in the original K-dimensional space.
%
% The selected features can be retrieved by W.DATA or by +W.
% See below for various methods to perform feature selection.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% MAPPINGS, DATASETS, FEATEVAL, FEATSELF, FEATSELLR,
% FEATSELO, FEATSELB, FEATSELI, FEATSELP, FEATSELM
% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands
function [w,v] = featsel(varargin)
if nargin == 2 && isscalar(varargin{1})
% preserving old type of call: [w,v] = featsel(k,j)
[k,j] = deal(varargin{:});
R = [1:k];
R(j) = [];
w = prmapping(mfilename,'fixed',j(:)',[],0,length(j));
v = prmapping(mfilename,'fixed',R(:)',[],0,length(R));
w = setname(w,'Feature Selection');
v = setname(v,'Feature Selection');
return
end
% new type of call: b = a*featsel(j)
argin = shiftargin(varargin,'vector');
argin = setdefaults(argin,[],[],false);
if mapping_task(argin,'definition')
w = define_mapping(argin,'fixed');
w = setname(w,'Feature Selection');
else
[a,j,inv] = deal(argin{:});
if ismapping(j)
j = +j;
end
if inv
R = [1:size(a,2)];
R(j) = [];
else
R = j;
end
w = a(:,R);
end