timeseries3.m

the matlab code of ginisvm - an svm implementation good at probability regression.

%-------------------------------------
% This script generates markovian data
% based on the input data and gaussian 
% distribution attached to each state.
%-------------------------------------

% 2 states
% 1 models

% Model 1
% Make sure that the sum of the probabilities = 1 #####
model1 = [ 0.4 0.3 0.3; 0.3 0.4 0.3; 0.3 0.3 0.4]';

% One mixture per state
weights = [ 1;1;1 ];

% Two dimensional data.
dim = 2;

% Parameters for each state
% and for each dimension ( mean and variance )
param1 = [ 0.0 0.2 0.0 0.2 ; 0.0 0.2 1 0.2; 1 0.2 0.0 0.2 ];

header = 200;
maxdata = 1000;

global poption;
global dynamics;
global startstate;
global finalstate;

poption = 0;
dynamics = 0;
startstate = 1;
finalstate = 4;

% Now to generate real data
% Each model can generate 
currtrain = 1;
currtest = 1;

[trainx,crossx,trainlabel,testlabel] = genmarkov(model1,weights,param1,dim,100,100);
traindata = trainx;
testdat = crossx;
[currtrain, dummy ] = size(trainx);
[currtest, dummy ] = size(crossx);

% store the test and the training data
fid = fopen('train.dat','w');
   for i = 1:currtrain,
      fprintf(fid,'%d\n',trainlabel(i)-1);
      for d = 1:dim,
         fprintf(fid,'%f\n',traindata(i,d));
      end;
      %fprintf(fid,'%f\n',traintime(i));
   end;
   fclose(fid);

% store the test and the training data
fid = fopen('test.dat','w');
   for i = 1:currtest,
      fprintf(fid,'%d\n',testlabel(i)-1);
      for d = 1:dim,
         fprintf(fid,'%f\n',testdat(i,d));
      end;
      %fprintf(fid,'%f\n',testtime(i));
   end;
   fclose(fid);

figure;
subplot(2,2,1);
hold on;
for i = 1:currtrain,
   if trainlabel(i) == 1,
      plot(traindata(i,1),traindata(i,2),'bo');
      Ytrain(i,:) = [ 1 0 0];
   else
      if trainlabel(i) == 2,
         plot(traindata(i,1),traindata(i,2),'r+');
         Ytrain(i,:) = [ 0 1 0];
      else
         plot(traindata(i,1),traindata(i,2),'g*');
         Ytrain(i,:) = [ 0 0 1];
      end;
   end;
end;
hold off;

subplot(2,2,2);
plot(trainlabel);
axis([0,100,0.7,2.3])

subplot(2,2,3);
hold on;
for i = 1:currtest,
   if testlabel(i) == 1,
      plot(testdat(i,1),testdat(i,2),'bo');
      Ycross(i,:) = [ 1 0 0];
   else
      if testlabel(i) == 2,
         plot(testdat(i,1),testdat(i,2),'r+');
         Ycross(i,:) = [ 0 1 0];
      else
         plot(testdat(i,1),testdat(i,2),'g*');
         Ycross(i,:) = [ 0 0 1];
      end;
   end;
end;
hold off;

subplot(2,2,4);
plot(testlabel);
axis([0,100,0.7,2.3])