Daniel@0: function [Class,P]=knn_old(Data, Proto, proto_class, K)
Daniel@0: 
Daniel@0: %KNN_OLD A K-nearest neighbor classifier using Euclidean distance 
Daniel@0: %
Daniel@0: % [Class,P]=knn_old(Data, Proto, proto_class, K)
Daniel@0: %
Daniel@0: %  [sM_class,P]=knn_old(sM, sData, [], 3);
Daniel@0: %  [sD_class,P]=knn_old(sD, sM, class);
Daniel@0: %  [class,P]=knn_old(data, proto, class);
Daniel@0: %  [class,P]=knn_old(sData, sM, class,5);
Daniel@0: %
Daniel@0: %  Input and output arguments ([]'s are optional): 
Daniel@0: %   Data   (matrix) size Nxd, vectors to be classified (=classifiees)
Daniel@0: %          (struct) map or data struct: map codebook vectors or
Daniel@0: %                   data vectors are considered as classifiees.
Daniel@0: %   Proto  (matrix) size Mxd, prototype vector matrix (=prototypes)
Daniel@0: %          (struct) map or data struct: map codebook vectors or
Daniel@0: %                   data vectors are considered as prototypes.
Daniel@0: %   [proto_class] (vector) size Nx1, integers 1,2,...,k indicating the
Daniel@0: %                   classes of corresponding protoptypes, default: see the 
Daniel@0: %                   explanation below. 
Daniel@0: %   [K]    (scalar) the K in KNN classifier, default is 1
Daniel@0: % 
Daniel@0: %   Class  (matrix) size Nx1, vector of 1,2, ..., k indicating the class 
Daniel@0: %                   desicion according to the KNN rule
Daniel@0: %   P      (matrix) size Nxk, the relative amount of prototypes of 
Daniel@0: %                   each class among the K closest prototypes for
Daniel@0: %                   each classifiee.
Daniel@0: %
Daniel@0: % If 'proto_class' is _not_ given, 'Proto' _must_ be a labeled SOM
Daniel@0: % Toolbox struct. The label of the data vector or the first label of
Daniel@0: % the map model vector is considered as class label for th prototype
Daniel@0: % vector. In this case the output 'Class' is a copy of 'Data' (map or
Daniel@0: % data struct) relabeled according to the classification.  If input
Daniel@0: % argument 'proto_class' _is_ given, the output argument 'Class' is
Daniel@0: % _always_ a vector of integers 1,2,...,k indiacating the class.
Daniel@0: %
Daniel@0: % If there is a tie between representatives of two or more classes
Daniel@0: % among the K closest neighbors to the classifiee, the class is
Daniel@0: % selected randomly among these candidates.
Daniel@0: %
Daniel@0: % IMPORTANT
Daniel@0: % 
Daniel@0: % ** Even if prototype vectors are given in a map struct the mask _is not 
Daniel@0: %    taken into account_ when calculating Euclidean distance
Daniel@0: % ** The function calculates the total distance matrix between all
Daniel@0: %    classifiees and prototype vectors. This results to an MxN matrix; 
Daniel@0: %    if N is high it is recommended to divide the matrix 'Data'
Daniel@0: %    (the classifiees) into smaller sets in order to avoid memory
Daniel@0: %    overflow or swapping. Also, if K>1 this function uses 'sort' which is
Daniel@0: %    considerably slower than 'max' which is used for K==1.
Daniel@0: %
Daniel@0: % See also KNN, SOM_LABEL, SOM_AUTOLABEL
Daniel@0: 
Daniel@0: % Contributed to SOM Toolbox 2.0, February 11th, 2000 by Johan Himberg
Daniel@0: % Copyright (c) by Johan Himberg
Daniel@0: % http://www.cis.hut.fi/projects/somtoolbox/
Daniel@0: 
Daniel@0: % Version 2.0beta Johan 040200
Daniel@0: 
Daniel@0: %% Init %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: % This must exist later
Daniel@0: classnames='';
Daniel@0: 
Daniel@0: % Check K 
Daniel@0: if nargin<4 | isempty(K),
Daniel@0:   K=1;
Daniel@0: end
Daniel@0: 
Daniel@0: if ~vis_valuetype(K,{'1x1'})
Daniel@0:   error('Value for K must be a scalar.');
Daniel@0: end
Daniel@0: 
Daniel@0: % Take data from data or map struct
Daniel@0: 
Daniel@0: if isstruct(Data);
Daniel@0:   if isfield(Data,'type') & ischar(Data.type),
Daniel@0:     ;
Daniel@0:   else
Daniel@0:     error('Invalid map/data struct?');
Daniel@0:   end
Daniel@0:   switch Data.type
Daniel@0:    case 'som_map'
Daniel@0:     data=Data.codebook;
Daniel@0:    case 'som_data'
Daniel@0:     data=Data.data;
Daniel@0:   end
Daniel@0: else
Daniel@0:   % is already a matrix
Daniel@0:   data=Data;
Daniel@0: end
Daniel@0: 
Daniel@0: % Take prototype vectors from prototype struct
Daniel@0: 
Daniel@0: if isstruct(Proto),
Daniel@0:   
Daniel@0:   if isfield(Proto,'type') & ischar(Proto.type),
Daniel@0:     ;
Daniel@0:   else
Daniel@0:     error('Invalid map/data struct?');
Daniel@0:   end
Daniel@0:   switch Proto.type
Daniel@0:    case 'som_map'
Daniel@0:     proto=Proto.codebook;
Daniel@0:    case 'som_data'
Daniel@0:     proto=Proto.data;
Daniel@0:   end
Daniel@0: else
Daniel@0:   % is already a matrix
Daniel@0:   proto=Proto; 
Daniel@0: end
Daniel@0: 
Daniel@0: % Check that inputs are matrices
Daniel@0: if ~vis_valuetype(proto,{'nxm'}) | ~vis_valuetype(data,{'nxm'}),
Daniel@0:   error('Prototype or data input not valid.')
Daniel@0: end
Daniel@0: 
Daniel@0: % Record data&proto sizes and check their dims 
Daniel@0: [N_data dim_data]=size(data); 
Daniel@0: [N_proto dim_proto]=size(proto);
Daniel@0: if dim_proto ~= dim_data,
Daniel@0:   error('Data and prototype vector dimension does not match.');
Daniel@0: end
Daniel@0: 
Daniel@0: % Check if the classes are given as labels (no class input arg.)
Daniel@0: % if they are take them from prototype struct
Daniel@0: 
Daniel@0: if nargin<3 | isempty(proto_class)
Daniel@0:   if ~isstruct(Proto)
Daniel@0:     error(['If prototypes are not in labeled map or data struct' ...
Daniel@0: 	   'class must be given.']);  
Daniel@0:     % transform to interger (numerical) class labels
Daniel@0:   else
Daniel@0:     [proto_class,classnames]=class2num(Proto.labels); 
Daniel@0:   end
Daniel@0: end
Daniel@0: 
Daniel@0: % Check class label vector: must be numerical and of integers
Daniel@0: if ~vis_valuetype(proto_class,{[N_proto 1]});
Daniel@0:   error(['Class vector is invalid: has to be a N-of-data_rows x 1' ...
Daniel@0: 	 ' vector of integers']);
Daniel@0: elseif sum(fix(proto_class)-proto_class)~=0
Daniel@0:   error('Class labels in vector ''Class'' must be integers.');
Daniel@0: end
Daniel@0: 
Daniel@0: % Find all class labels
Daniel@0: ClassIndex=unique(proto_class);
Daniel@0: N_class=length(ClassIndex); % number of different classes  
Daniel@0: 
Daniel@0: % Calculate euclidean distances between classifiees and prototypes
Daniel@0: d=distance(proto,data);
Daniel@0: 
Daniel@0: %%%% Classification %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: 
Daniel@0: if K==1,   % sort distances only if K>1
Daniel@0:   
Daniel@0:   % 1NN
Daniel@0:   % Select the closest prototype
Daniel@0:   [tmp,proto_index]=min(d);
Daniel@0:   class=proto_class(proto_index);
Daniel@0: 
Daniel@0: else 
Daniel@0:   
Daniel@0:   % Sort the prototypes for each classifiee according to distance
Daniel@0:   [tmp,proto_index]=sort(d);
Daniel@0:   
Daniel@0:   %% Select K closest prototypes
Daniel@0:   proto_index=proto_index(1:K,:);
Daniel@0:   knn_class=proto_class(proto_index);
Daniel@0:   for i=1:N_class,
Daniel@0:     classcounter(i,:)=sum(knn_class==ClassIndex(i));
Daniel@0:   end
Daniel@0:   
Daniel@0:   %% Vote between classes of K neighbors 
Daniel@0:   [winner,vote_index]=max(classcounter);
Daniel@0:   
Daniel@0:   %% Handle ties
Daniel@0:   
Daniel@0:   % set index to clases that got as amuch votes as winner
Daniel@0:   
Daniel@0:   equal_to_winner=(repmat(winner,N_class,1)==classcounter);
Daniel@0:   
Daniel@0:   % set index to ties
Daniel@0:   tie_index=find(sum(equal_to_winner)>1); % drop the winner from counter 
Daniel@0:   
Daniel@0:   % Go through equal classes and reset vote_index randomly to one
Daniel@0:   % of them 
Daniel@0:   
Daniel@0:   for i=1:length(tie_index),
Daniel@0:     tie_class_index=find(equal_to_winner(:,tie_index(i)));
Daniel@0:     fortuna=randperm(length(tie_class_index));
Daniel@0:     vote_index(tie_index(i))=tie_class_index(fortuna(1));
Daniel@0:   end
Daniel@0:   
Daniel@0:   class=ClassIndex(vote_index);
Daniel@0: end
Daniel@0: 
Daniel@0: %% Build output %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: 
Daniel@0: % Relative amount of classes in K neighbors for each classifiee
Daniel@0: 
Daniel@0: if K==1,
Daniel@0:   P=zeros(N_data,N_class);
Daniel@0:   if nargout>1,
Daniel@0:     for i=1:N_data,
Daniel@0:       P(i,ClassIndex==class(i))=1;
Daniel@0:     end
Daniel@0:   end
Daniel@0: else
Daniel@0:   P=classcounter'./K;
Daniel@0: end
Daniel@0: 
Daniel@0: % xMake class names to struct if they exist
Daniel@0: if ~isempty(classnames),
Daniel@0:   Class=Data;
Daniel@0:   for i=1:N_data,
Daniel@0:     Class.labels{i,1}=classnames{class(i)};
Daniel@0:   end
Daniel@0: else
Daniel@0:   Class=class;
Daniel@0: end
Daniel@0: 
Daniel@0: 
Daniel@0: %%% Subfunctions %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%  
Daniel@0: 
Daniel@0: function [nos,names] = class2num(class)
Daniel@0: 
Daniel@0: % Change string labels in map/data struct to integer numbers
Daniel@0: 
Daniel@0: names = {};
Daniel@0: nos = zeros(length(class),1);
Daniel@0: for i=1:length(class)
Daniel@0:   if ~isempty(class{i}) & ~any(strcmp(class{i},names))
Daniel@0:     names=cat(1,names,class(i));
Daniel@0:   end
Daniel@0: end
Daniel@0: 
Daniel@0: tmp_nos = (1:length(names))';
Daniel@0: for i=1:length(class)
Daniel@0:   if ~isempty(class{i})
Daniel@0:     nos(i,1) = find(strcmp(class{i},names));    
Daniel@0:   end
Daniel@0: end
Daniel@0: 
Daniel@0: function d=distance(X,Y);
Daniel@0: 
Daniel@0: % Euclidean distance matrix between row vectors in X and Y
Daniel@0: 
Daniel@0: U=~isnan(Y); Y(~U)=0;
Daniel@0: V=~isnan(X); X(~V)=0;
Daniel@0: d=X.^2*U'+V*Y'.^2-2*X*Y';