lg_cluster.m

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

function model = LG_Cluster(X_in,lrn_par)

% function model = LG_Cluster(X_in,lrn_par)
%
% Build a clustering model
%   returns a model
% X_in is the unlabeled input data (one case per row)
% lrn_par are parameters - see Set_Default_Learning_Paramters
% returns a model
%
%
% 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.
%

USE_C_CODE_FOR_DST = lrn_par.USE_C_CODE_FOR_DST;

[n,dim] = size(X_in);

% Scaling
model.DST_TYPE = lrn_par.DST_TYPE;
model.SCALE = lrn_par.SCALE;
if lrn_par.DST_TYPE == 1 % Euclidean
    if model.SCALE == 1
        [X, A, B] = Gen_Scale(X_in,0,1);
        model.A = A;
        model.B = B;
    else
        X = X_in;
    end
else % Dot Product dist
    if model.SCALE == 1
        model.B = mean(X_in);
        X_in_2 = X_in - model.B(ones(n,1),:);
        x_row_ln = (sum(X_in_2'.^2).^(-1/2))';
        X = X_in_2.*(x_row_ln(:,ones(dim,1)));
    else
        x_row_ln = (sum(X_in'.^2).^(-1/2))';
        X = X_in.*(x_row_ln(:,ones(dim,1)));
    end
    %     (sum(X'.^2).^(-1/2))'
    %     t99 = 1;
end
clear X_in;


% Distance Matrix
if lrn_par.DST_SAVED == 1
    load dst.mat;
else
    if lrn_par.DST_TYPE  == 1
        if USE_C_CODE_FOR_DST == 1
            dst = distance_matrix_mex(dim,n,X');
        else
            dst = zeros(n,n);
            for i=2:n
                j=[1:i-1];
                t1 = X(i*ones(i-1,1),:) - X(j,:); %% distance function
                dst(i+[j-1]*n) = sum(t1.*t1,2);
                dst(j+(i-1)*n) = dst(i+[j-1]*n);
            end
        end
    else
        dst = 1 - X*X';
    end
    save dst.mat dst
end

if lrn_par.DST_TYPE == 1
    dim_it = dim;
else
    dim_it = 1;
end

fprintf(1,'Now opt_alpha_sigma...\n');

if lrn_par.my_alpha ~= 0 & lrn_par.my_sigma ~= 0 & length(lrn_par.Cluster_Search) == 1
    my_alpha = lrn_par.my_alpha;
    my_sigma = lrn_par.my_sigma;
    num_classes = lrn_par.Cluster_Search;
else
    [my_alpha,my_sigma,num_classes] = Opt_Alpha_Sigma(dst,lrn_par,dim_it,n);
end

fprintf(1,'sigma %f alpha %f nc %i\n',my_sigma,my_alpha,num_classes);
fprintf(1,'Now Calc_Inv...\n');

[iaS_inv, norm_iaS_inv, S_norm, F_orig, F_norm, Y, Y_orig, dot_dist, mean_dist, ind_mat_unseen, Dt, ind_clust] =...
    Calc_Inv(dst,my_alpha,my_sigma,num_classes,n,lrn_par);

num_classes = length(ind_clust);

Class_Outlier = [];
for i=1:num_classes
    ind_t = logical(Y(:,i));
    ind_vals = find(Y(:,i) == 1);
    t1 = norm_iaS_inv(ind_t,ind_t);
    t2 = Gen_Scale(t1,0,1);
    len_cols = sum(t2.^2).^(-0.5);
    t3 = t2.*(len_cols(ones(sum(Y(:,i)),1),:));
    t4 = t3' * t3;
    t5 = mean(t4);
    Class_Outlier(i).val = t5';
    Class_Outlier(i).ind = ind_t;
    [s_val,ind_sort] = sort(t5);
    Class_Outlier(i).sort_val = s_val;
    Class_Outlier(i).ind_sort = ind_vals(ind_sort);
    plot_values = zeros(size(Y(:,i)));
    plot_values(ind_t) = t5';
    Class_Outlier(i).plot_values = plot_values;
    %figure;plot(Class_Outlier(i).plot_values,'r*'); %plot the points in each cluster
end

model.Class_Outlier = Class_Outlier;

model.F_orig = F_orig;
model.F_norm = F_norm;
model.Y = Y;
model.Y_orig = Y_orig;
model.mean_dist = mean_dist;

[sort_mean_dist,ind_sort_mean_dist] = sort(mean_dist);
model.sort_mean_dist = sort_mean_dist;
model.ind_sort_mean_dist = ind_sort_mean_dist;


model.S_norm = S_norm;
model.ind_mat_unseen = ind_mat_unseen;
one_over_2_sigma_sq = 1/(2*my_sigma^2);
model.one_over_2_sigma_sq = one_over_2_sigma_sq;
model.X = X;
model.Dt = Dt;

model.num_classes = num_classes;

model.norm_iaS_inv = norm_iaS_inv;

model.iaS_inv = iaS_inv;

model.my_alpha = my_alpha;
model.my_sigma = my_sigma;

model.ind_clust = ind_clust;

t2 = 99;