nndd.m

模式识别工具箱,希望对大家有用!

%NNDD Nearest neighbour data description method.
% 
%       W = nndd(A,fracrej)
% 
% Calculates the Nearest neighbour data description. Training only
% consists of the computation of the resemblance of all training
% objects to the training data using Leave-one-out.
% 
% See also datasets, mappings, dd_roc

% Copyright: D. Tax, R.P.W. Duin, [e-mail]
% Faculty of Applied Physics, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands
  
function [W,out] = nndd(a,fracrej)

if nargin < 2 | isempty(fracrej), fracrej = 0.05; end
if nargin < 1 | isempty(a) % empty nndd
  W = mapping(mfilename,{fracrej});
  return
end

if isa(fracrej,'double')           %training

  if isa(a,'dataset')              %train on training set
    a = target_class(a);      % make sure we have a OneClass dataset
    [nlab,lablist,m,k,c] = dataset(a);

    % settings:
    small_D = 1.0e-10;  % small distance,
    large_D = 1.0e+10;  % large distance.

    % leave-one-out on the training set:
    fit = zeros(m,1);

    distmat = distm(a);
    distmat = distmat + large_D*(distmat<small_D); %surpress 0 dist.
    for i=1:m
      D = distmat;
      [minD minI] = min(D(i,:));  % dist. from z to 1NN in A
      D(i,minI) = large_D;
      intdist = min(D(:,minI));   % dist. from 1NN to NN(1NN)
      fit(i) = minD./intdist;
    end
  else
    % check that a is now an  nndd?
    [W,classlist,type,k,c] = mapping(a);
    fit = W.fit;
  end
  %now obtain the threshold:
  thresh = -threshold(-fit,fracrej);
  %and save all useful data:
  out = fit;
  W.x = +a;
  W.threshold = thresh;
  W.fit = fit;
  W.D = min(distmat,[],2);
  W.scale = mean(fit);
%  W = {+a,thresh,fit,min(distmat,[],2),mean(fit)};
  W = mapping(mfilename,W,str2mat('target','outlier'),k,c);

else                               %testing

  [W,classlist,type,k,c] = mapping(fracrej);  % unpack
  [nlab,lablist,m,k,c,p] = dataset(a);

  %compute:
  distmat = distm(a,W.x);    %dist between train and test
  [mindist I] = min(distmat,[],2);
  out = log([mindist./(W.D(I)) ones(m,1)*W.threshold]);

  % map to probability
  newout = dist2dens(out,W.scale);
           
  W = dataset(newout,getlab(a),classlist,p,lablist);
end
return