batch_test_toys.m

来自「clustering_code ,Clustering Through Rank」· M 代码 · 共 327 行

function test_toys

%
% 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.
%

close all;
dbstop if error;
dbstop if warning;

rand('seed',7654321);

H=figure;

fprintf(1,'2 moons\n');
load Moon_Data.mat;
Data_X=Gen_Scale(Data_X,0,1);
[Data_X_us,True_Y_us]=getTestData(Data_X,True_Y);
figure; scatter(Data_X_us(:,1),Data_X_us(:,2),20,True_Y_us*(1:2)');
test(Data_X,True_Y,Data_X_us,True_Y_us,0.1,2 ,H,1);

fprintf(1,'Spiral\n');
[Data_X2,True_Y2]=getSpiral;
[Data_X_us2,True_Y_us2]=getTestData(Data_X2,True_Y2);
figure; scatter3(Data_X_us2(:,1),Data_X_us2(:,2),Data_X_us2(:,3),20,True_Y_us2*(1:3)');
test(Data_X2,True_Y2,Data_X_us2,True_Y_us2,0.1,3 ,H,2);

%save toy_stuff_spectral.mat

function test(Data_X,True_Y,Data_X_us,True_Y_us,scale,clusters, figureHandle, subplotRow)

maxRows=2;
maxCols=2;

%% CHANGE STUFF HERE
neighbor_num=3;
nGroups=clusters;
X=Data_X;
clear Data_X;

    %% centralize and scale the data
    X = X - repmat(mean(X),size(X,1),1);
    X = X/max(max(abs(X)));
    Data_X_us = Data_X_us - repmat(mean(Data_X_us),size(Data_X_us,1),1);
    Data_X_us = Data_X_us/max(max(abs(Data_X_us)));

lrn_par = Set_Default_Learning_Paramters;
% Optimization for both ALPHA and SIGMA together
lrn_par.OPT_S_A = 1; % set to 1 to optimize for both alpha and sigma
lrn_par.SIGMA_MEDIAN = 0;
lrn_par.Cluster_Search = [2,3,4]; % This defines the clusters to be evaluated
lrn_par.Cluster_Search = clusters;
lrn_par.my_sigma = 0;%scale;
lrn_par.my_alpha = 0;

model = LG_Cluster(X,lrn_par);

error_thing(True_Y,model.Y);
[F_norm,FA,FB] = Gen_Scale(model.F_norm.^2,0.1,1);

figure(figureHandle);
subplot(maxRows,maxCols, (subplotRow-1)*maxRows+1);
hold on;
if size(X,2)==2
    %scatter(X(:,1),X(:,2),20*F_norm(model.Y*(1:nGroups)'),model.Y*(1:nGroups)');
    %scatter(X(model.ind_clust,1),X(model.ind_clust,2),100,model.Y(model.ind_clust,:)*(nGroups+(1:nGroups))','*');
    labels=model.Y*(1:nGroups)';
    ind = find(labels==1);
    plot3(X(ind,1),X(ind,2),(model.Class_Outlier(1).val)*2-1,'bx','MarkerSize',10);
    plot3(X(model.ind_clust(1),1),X(model.ind_clust(1),2),max(model.Class_Outlier(1).val),'k*','MarkerSize',20);
    ind = find(labels==2);
    plot3(X(ind,1),X(ind,2),(model.Class_Outlier(2).val)*2-1,'r+','MarkerSize',10);
    plot3(X(model.ind_clust(2),1),X(model.ind_clust(2),2),max(model.Class_Outlier(2).val),'k*','MarkerSize',20);
    view(45,50);
    xlabel('X_1','FontSize',14);
    ylabel('X_2','FontSize',14);
    zlabel('Outlier Measure','FontSize',14);
else
    scatter3(X(:,1),X(:,2),X(:,3),200*F_norm(model.Y*(1:nGroups)'),model.Y*(1:nGroups)');
    scatter3(X(model.ind_clust,1),X(model.ind_clust,2),X(model.ind_clust,3),100,model.Y(model.ind_clust,:)*(nGroups+(1:nGroups))','*');
    view(45,50);
end;
title(sprintf('Clustering sigma=%.2f',model.my_sigma),'FontSize',18);

%% now try unseen digits
fprintf(1,'Now unseen data...\n\n\n');

%Y=mapSpectralOutOfSample(Data_X_us,diag(misc.ss(1:nGroups,1:nGroups)),misc.V(:,1:nGroups),X,scale,nGroups,misc.mu);

[un_class] = Classify_New_Data(Data_X_us,model);

error_thing(True_Y_us,un_class.Y);
F_norm=(un_class.F.^2) .* repmat(FA',size(un_class.F,1),1)+repmat(FB',size(un_class.F,1),1);

figure(figureHandle);
subplot(maxRows,maxCols, (subplotRow-1)*maxRows+2);
if size(X,2)==2
    scatter(Data_X_us(:,1),Data_X_us(:,2),20*F_norm(un_class.Y*(1:nGroups)'),un_class.Y*(1:nGroups)');
else
    scatter3(Data_X_us(:,1),Data_X_us(:,2),Data_X_us(:,3),200*F_norm(un_class.Y*(1:nGroups)'),un_class.Y*(1:nGroups)');
end;
title('Out-of-sample data','FontSize',14);

return;

function [X,Y]=getTestData(X,Y)
    distort=0.2;
    X=X+2*distort*rand(size(X))-distort;
    X(find(X<0))=0;
    X(find(X>1))=1;    
return;

function error_thing(True_Y,Y)
    nGroups = size(True_Y,2);
    tab=crosstab(True_Y*(1:nGroups)',Y*(1:nGroups)');
    xmax=max(tab);
    dcount=1;
    for i=1:length(xmax) 
        xx=find(tab(:,i)==xmax(i));
        if length(xx)>1
            myperm(i)=xx(dcount);
            dcount=dcount+1;
        else
            myperm(i)=xx(1);
        end;
    end;
    for i=1:length(myperm)
        xx=find(myperm==i);
        if length(xx)<1
            xmyperm(i)=1;
        else
            xmyperm(i)=xx(1);
        end;
    end;
    %myperm=xmyperm;
    tab=[tab(:,xmyperm), tab]

    if length(myperm)~=nGroups
        myperm = 1:nGroups;
        fprintf(1,'Error with the error thing\n');
    end;
    True_Y=True_Y(:,myperm);
    
    errors = abs(Y - True_Y);
    err_ind = find(sum(abs(Y - True_Y)')'>0);
    num_errors = length(err_ind);
    err_rate = num_errors / length(True_Y)
return;


function [X,Y]=getSpiral

deg = 0:0.05:2*pi;
rad_min = 0.95;
rad_max = 1.0;
X1 = zeros(length(deg),3);
for i=1:length(deg)
    r = rad_min + rand(1,1)*(rad_max-rad_min);
    X1(i,1) = r * cos(deg(i));
    X1(i,2) = r * sin(deg(i));
    X1(i,3) = (i/length(deg));
end


X2 = zeros(length(deg),3);
for i=1:length(deg)
    r = rad_min + rand(1,1)*(rad_max-rad_min);
    X2(i,1) = r * cos(deg(i) + (2 * pi/3));
    X2(i,2) = r * sin(deg(i) + (2 * pi/3));
    X2(i,3) = (i/length(deg));
end

X3 = zeros(length(deg),3);
for i=1:length(deg)
    r = rad_min + rand(1,1)*(rad_max-rad_min);
    X3(i,1) = r * cos(deg(i) + (4 * pi/3));
    X3(i,2) = r * sin(deg(i) + (4 * pi/3));
    X3(i,3) = (i/length(deg));
end

X1=Gen_Scale(X1,0,1);
X2=Gen_Scale(X2,0,1);
X3=Gen_Scale(X3,0,1);

figure;
hold on
plot3(X1(:,1),X1(:,2),X1(:,3),'r+');
plot3(X2(:,1),X2(:,2),X2(:,3),'g*');
plot3(X3(:,1),X3(:,2),X3(:,3),'bs');
hold off
pause(1);
t99 = 0;

X = [X1;X2;X3];
[num_ex,dim] = size(X);

Y=eye(3);
Y=Y([ones(length(X1),1),2*ones(length(X2),1),3*ones(length(X3),1)],:);

return;

% >> 2 moons
%  
%  Func-count     x          f(x)         Procedure
%     1      0.0543197    -0.914932        initial
%     2      0.0566803    -0.909387        golden
%     3      0.0528607    -0.915232        golden
%     4       0.053407    -0.915372        parabolic
%     5      0.0533737    -0.915372        parabolic
%     6      0.0534404    -0.915371        parabolic
% 
% Optimization terminated successfully:
%  the current x satisfies the termination criteria using OPTIONS.TolX of 1.000000e-004 
% 
% 
%                                max                   Directional   First-order 
%  Iter F-count        f(x)   constraint    Step-size   derivative   optimality Procedure 
%     1      8    -0.963494       -0.005          0.5      -0.0952        0.868   
%     2     13    -0.985601      -0.0025          0.5      -0.0345         3.33   
%     3     18    -0.994939     -0.00125          0.5      -0.0152         2.69   
%     4     23    -0.996957    -0.000625          0.5     0.000125         5.27   
%     5     28    -0.999199   -0.0003125          0.5     -0.00253         3.37   
%     6     33    -0.999458   -0.0001562          0.5    -0.000116         2.28   
%     7     37    -0.999875  -5.868e-005            1    -0.000171         1.03   
%     8     41    -0.999921  -7.718e-005            1    -1.4e-005        0.572  Hessian modified  
%     9     45    -0.999973  -2.487e-005            1   -1.45e-005        0.456  Hessian modified  
%    10     49    -0.999984  -1.702e-005            1   -8.36e-006        0.442  Hessian modified  
%    11     53    -0.999993  -1.751e-005            1   -4.22e-006        0.171  Hessian modified  
%    12     57    -0.999996    -3.5e-006            1    7.43e-006        0.981  Hessian modified  
%    13     61    -0.999997  -9.909e-006            1    8.22e-007        0.141  Hessian modified  
% Optimization terminated successfully:
%  Magnitude of directional derivative in search direction 
%   less than 2*options.TolFun and maximum constraint violation 
%   is less than options.TolCon
%  No Active Constraints
% 
% tab =
% 
%    105     0   105     0
%      0   105     0   105
% 
% 
% err_rate =
% 
%      0
% 
% Now unseen data...
% 
% 
% 
% tab =
% 
%    105     0   105     0
%      0   105     0   105
% 
% 
% err_rate =
% 
%      0
% 
% Spiral
%  
%  Func-count     x          f(x)         Procedure
%     1      0.0443197    -0.522318        initial
%     2      0.0466803    -0.512971        golden
%     3      0.0428607    -0.520102        golden
%     4      0.0441196     -0.52232        parabolic
%     5      0.0442159    -0.522332        parabolic
%     6      0.0442493    -0.522331        parabolic
%     7      0.0441826    -0.522331        parabolic
% 
% Optimization terminated successfully:
%  the current x satisfies the termination criteria using OPTIONS.TolX of 1.000000e-004 
% 
% 
%                                max                   Directional   First-order 
%  Iter F-count        f(x)   constraint    Step-size   derivative   optimality Procedure 
%     1      8    -0.626496       -0.005          0.5        -0.23         15.2   
%     2     13    -0.675386      -0.0025          0.5       -0.014         42.8  Hessian modified  
%     3     18    -0.785077     -0.00125          0.5       -0.198         46.3   
%     4     23    -0.843684    -0.000625          0.5        0.195         87.2   
%     5     27    -0.999999            0            1        0.029          251   
%     6     48           -1  -2.984e-007    7.63e-006       0.0233         1.78   
%     7     52           -1  -2.946e-007            1   -2.23e-009        0.583   
% Optimization terminated successfully:
%  Search direction less than 2*options.TolX and
%   maximum constraint violation is less than options.TolCon
%  No Active Constraints
% 
% tab =
% 
%    126     0     0     0   126     0
%      0   126     0   126     0     0
%      0     0   126     0     0   126
% 
% 
% err_rate =
% 
%      0
% 
% Now unseen data...
% 
% 
% 
% tab =
% 
%    126     0     0     0   126     0
%      0   126     0   126     0     0
%      0     0   126     0     0   126
% 
% 
% err_rate =
% 
%      0