classify_new_data.m

clustering_code ,Clustering Through Ranking On Manifolds Version 0.2 Copyright by Markus Breit

function [us_class] = Classify_New_Data(X_in,model)

% [us_class] = Classify_New_Data(X_in,model)
% Classify unseen examples - X_in is the input data (one case per row)

%
% Clustering Through Ranking On Manifolds
% Version 0.2
%
% Copyright by Markus Breitenbach and Gregory Z. Grudic
% This code is for your personal and research use only.
%
% http://www.cs.colorado.edu/~grudic/
% http://ucsu.colorado.edu/~breitenm/
%
% This software is provided "as is," without warranty of any kind, express
% or implied.  In no event shall the authors be held liable
% for any direct, indirect, incidental, special or consequential damages
% arising out of the use of or inability to use this software.
%

[num_data,d] = size(X_in);
if model.DST_TYPE == 1
    if model.SCALE == 1
        A = model.A';
        B = model.B';
        X =  X_in.*A(ones(num_data,1),:) + B(ones(num_data,1),:);
        X(X>1) = 1;
        X(X<0) = 0;
    else
        X = X_in;
    end
else
    if model.SCALE == 1
        model.B = mean(X_in);
        X_in_2 = X_in - model.B(ones(num_data,1),:);
        x_row_ln = (sum(X_in_2'.^2).^(-1/2))';
        X = X_in_2.*(x_row_ln(:,ones(d,1)));
    else
        x_row_ln = (sum(X_in'.^2).^(-1/2))';
        X = X_in.*(x_row_ln(:,ones(d,1)));
        t77 = 77;
    end
end


Data_X = model.X;
one_over_2_sigma_sq = model.one_over_2_sigma_sq;
S_norm = model.S_norm;
ind_mat_unseen = model.ind_mat_unseen;
Dt = model.Dt;

% model.ind_clust = ind_clust;
% model.num_models = num_models;
% model.Dt = Dt;

[n,dim] = size(Data_X);
ind_diag = find(eye(n,n) == 1);

coeff_unseen = zeros(num_data,n);
iaS_inv_un_seen = zeros(n,num_data);
outlier_est_unseen = zeros(n,1);


% model.S_norm = S_norm(ind_clust,:);
% model.ind_mat_unseen = ind_mat_unseen(ind_clust,:);
% model.one_over_2_sigma_sq = one_over_2_sigma_sq;
% model.X_clust = X(ind_clust,:);


UNS_one_over_2_sigma_sq = 1/(2*0.05^2);


for pt = 1:num_data
    
    %    dst_us = zeros(1,n);
    %     for i=1:n
    %         t1 = Data_X(i,:) - X(pt,:);
    %         dst_us(i) = t1*t1';
    %     end
    if model.DST_TYPE == 1
        %t1 = Data_X - X(pt*ones(n,1),:);
        %dst_us = sum(t1.*t1,2);
        if pt==1
            ncentres = size(Data_X,1);
            ndata    = size(X,1);
            all_my_distances= (ones(ncentres, 1) * sum((X.^2)', 1))' + ones(ndata, 1) * sum((Data_X.^2)',1) - 2.*(X*(Data_X'));
        end;
        dst_us = all_my_distances(pt,:);
    else
        dst_us = 1 - Data_X*X(pt*ones(n,1),:)';
        t99 = 0;
    end

    
    c_us = exp(-dst_us.*one_over_2_sigma_sq);
    
    W_us = c_us(ind_mat_unseen);
    W_us(ind_diag) = 0;
   
    
    Dt_us = sum(W_us').^(-1/2);
    %S_us = diag(Dt_us) * W_us * diag(Dt);
    S_us = (W_us.*Dt_us(ones(n,1),:)').*((Dt(ones(n,1),:)));

    
    len_rows = sum(S_us'.^2).^(-0.5);
    S_norm_us = S_us.*(len_rows(ones(n,1),:))';

    
    t2 = sum((S_norm_us.*S_norm)');
    t22 = (t2/(sum(t2)));
    %     t_d = 1-t2;
    %     t_d_s = Gen_Scale(t_d',0,1);
    %     t_scale = 1-t_d_s;
    %     t22 = (t2.*t_scale').^2;
    %t22 = exp(-(1-t2).*UNS_one_over_2_sigma_sq);
    coeff_unseen(pt,:) = t22;
    t3 = model.norm_iaS_inv.*t22(ones(n,1),:);
    iaS_inv_un_seen(:,pt) = sum(t3')';
    t99 = 8;
end

len_cols = sum(iaS_inv_un_seen.^2).^(-0.5);
norm_iaS_inv_un_seen = iaS_inv_un_seen.*(len_cols(ones(n,1),:));

dot_dist = norm_iaS_inv_un_seen' * model.norm_iaS_inv;
us_class.outlier_est_unseen = mean(dot_dist');

F_us_nn = norm_iaS_inv_un_seen' * model.Y;
len_cols_F_us_nn = sum(F_us_nn.^2).^(-0.5);
F_us = F_us_nn.*(len_cols_F_us_nn(ones(num_data,1),:));

% Fnorm_us = model.norm_iaS_inv * model.Y;
% len_rows_F = sum(Fnorm_us.^2).^(-0.5);
% Fnorm_us_norm = Fnorm_us.*(len_rows_F(ones(n,1),:));
% 
% len_rows = sum(coeff_unseen').^(-1);
% coeff_unseen = coeff_unseen.*(len_rows(ones(n,1),:))';
% 
% 
% F_us_nn = coeff_unseen * Fnorm_us_norm;
% len_cols_F_us_nn = sum(F_us_nn.^2).^(-0.5);
% F_us = F_us_nn.*(len_cols_F_us_nn(ones(n,1),:));
% 
% 
% dot_dist = coeff_unseen * model.norm_iaS_inv;
% us_class.outlier_est_unseen = mean(dot_dist);


Y = zeros(size(F_us));
for i=1:num_data
    ind_t = find(max(F_us(i,:)) == F_us(i,:));
    Y(i,ind_t) = 1;
end

Class_Outlier = [];
for i=1:model.num_classes
    ind_t = logical(Y(:,i));
    ind_vals = find(Y(:,i) == 1);
    scale_mat = coeff_unseen(ind_t,model.Class_Outlier(i).ind);
    %len_rows4 = sum(scale_mat'.^2).^(-0.5);
    len_rows4 = sum(scale_mat').^(-1);
    scale_mat_norm = scale_mat.*(len_rows4(ones(sum(model.Class_Outlier(i).ind),1),:))';
    
    t1 = scale_mat_norm * model.Class_Outlier(i).val;

    Class_Outlier(i).val = t1;
    plot_values = zeros(size(F_us(:,i)));
    plot_values(ind_t) = t1;
    Class_Outlier(i).plot_values = plot_values;
    figure;plot(Class_Outlier(i).plot_values,'b^'); %plot the points in each cluster
end

us_class.F = F_us;
us_class.Y = Y;
us_class.Class_Outlier = Class_Outlier;
us_class.Fs= Gen_Scale(F_us,0.1,1);

t99 = 7;