Sunday 28 October 2018

mnistdeepauto analysis 6

%backprop.m
% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.

% This program fine-tunes an autoencoder with backpropagation.
% Weights of the autoencoder are going to be saved in mnist_weights.mat
% and trainig and test reconstruction errors in mnist_error.mat
% You can also set maxepoch, default value is 200 as in our paper. 

maxepoch=1; %maxepoch=200;
fprintf(1,'\nFine-tuning deep autoencoder by minimizing cross entropy error. \n');
fprintf(1,'60 batches of 1000 cases each. \n');

load mnistvh
load mnisthp
load mnisthp2
load mnistpo

makebatches;
[numcases numdims numbatches]=size(batchdata); % 100x784x600
N=numcases;

%%%% PREINITIALIZE WEIGHTS OF THE AUTOENCODER %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
w1=[vishid; hidrecbiases]; % 784x1000 ; 1x1000 = 785x1000
w2=[hidpen; penrecbiases]; % 1000x500 ; 1x500 = 1001x500
w3=[hidpen2; penrecbiases2]; %500x250 ; 1x250 = 501x250
w4=[hidtop; toprecbiases]; % 250x30 ; 1x30 = 251x30
w5=[hidtop'; topgenbiases]; % 30x250 ; 1x250 = 31x250
w6=[hidpen2'; hidgenbiases2]; % 250x500 ; 1x500 = 251x500
w7=[hidpen'; hidgenbiases]; % 500x1000 ; 1x1000 = 501x1000
w8=[vishid'; visbiases]; % 1000x784 ; 1x784 = 1001x784

%%%%%%%%%% END OF PREINITIALIZATIO OF WEIGHTS  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

l1=size(w1,1)-1; % 784
l2=size(w2,1)-1; % 1000
l3=size(w3,1)-1; % 500
l4=size(w4,1)-1; % 250
l5=size(w5,1)-1; % 30
l6=size(w6,1)-1; % 250
l7=size(w7,1)-1; % 500
l8=size(w8,1)-1; % 1000
l9=l1;  % 784
test_err=[];
train_err=[];

for epoch = 1:maxepoch

 %%%%%%%%%%%%%%%%%%%% COMPUTE TRAINING RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 err=0;
 [numcases numdims numbatches]=size(batchdata); % 100x784x600
 N=numcases; % 100
 for batch = 1:numbatches
  data = [batchdata(:,:,batch)]; % 100x784
  data = [data ones(N,1)]; % 100x785 = 100x784 ++ 100x1
  w1probs = 1./(1 + exp(-data*w1)); w1probs = [w1probs  ones(N,1)]; % 100x785 * 785x1000 ++ 100x1001
  w2probs = 1./(1 + exp(-w1probs*w2)); w2probs = [w2probs ones(N,1)]; % 100x1001 * 1001x500 ++ 100x501
  w3probs = 1./(1 + exp(-w2probs*w3)); w3probs = [w3probs  ones(N,1)]; % 100x501 * 501x250 ++ 100x251
  w4probs = w3probs*w4; w4probs = [w4probs  ones(N,1)]; % 100x251 * 251x30 ++ 100x31
  w5probs = 1./(1 + exp(-w4probs*w5)); w5probs = [w5probs  ones(N,1)]; % 100x31 * 31x250 ++ 100x251
  w6probs = 1./(1 + exp(-w5probs*w6)); w6probs = [w6probs  ones(N,1)]; % 100x251 * 251x500 ++ 100x501
  w7probs = 1./(1 + exp(-w6probs*w7)); w7probs = [w7probs  ones(N,1)]; %  100x501 * 501x1000 ++ 100x1001
  dataout = 1./(1 + exp(-w7probs*w8)); % 100x1001 * 1001x784 = 100x784
  err= err +  1/N*sum(sum( (data(:,1:end-1)-dataout).^2 )); % 100x784 - 100x784
 end
 train_err(epoch)=err/numbatches;

 %%%%%%%%%%%%%% END OF COMPUTING TRAINING RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%

 %%%% DISPLAY FIGURE TOP ROW REAL DATA BOTTOM ROW RECONSTRUCTIONS %%%%%%%%%%%%%%%%%%%%%%%%%
 fprintf(1,'Displaying in figure 1: Top row - real data, Bottom row -- reconstructions \n');
 output=[];
 for ii=1:15
  output = [output data(ii,1:end-1)' dataout(ii,:)']; % 784x100 ++ 784x1
 end
   if epoch==1
   close all
   figure('Position',[100,600,1000,200]);
   else
   figure(1)
   end
   mnistdisp(output);
   drawnow;

 %%%%%%%%%%%%%%%%%%%% COMPUTE TEST RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 [testnumcases testnumdims testnumbatches]=size(testbatchdata); % 100x784x100
 N=testnumcases; % 100
 err=0;
 for batch = 1:testnumbatches
  data = [testbatchdata(:,:,batch)];
  data = [data ones(N,1)]; % 100x785
  w1probs = 1./(1 + exp(-data*w1)); w1probs = [w1probs  ones(N,1)]; %100x785 * 785x1000 -> 100x1001
  w2probs = 1./(1 + exp(-w1probs*w2)); w2probs = [w2probs ones(N,1)]; % 100x1001 * 1001x500 -> 1001x501
  w3probs = 1./(1 + exp(-w2probs*w3)); w3probs = [w3probs  ones(N,1)]; % 100x501 * 501x250 -> 100x251
  w4probs = w3probs*w4; w4probs = [w4probs  ones(N,1)]; % 100x251 * 251x30 -> 100x31
  w5probs = 1./(1 + exp(-w4probs*w5)); w5probs = [w5probs  ones(N,1)]; % 100x31 * 31x250 -> 100x251
  w6probs = 1./(1 + exp(-w5probs*w6)); w6probs = [w6probs  ones(N,1)]; % 100x251 * 251x500 -> 100x501
  w7probs = 1./(1 + exp(-w6probs*w7)); w7probs = [w7probs  ones(N,1)]; % 100x501 * 501x1000 -> 100x1001
  dataout = 1./(1 + exp(-w7probs*w8)); % 100x1001 * 1001x784 = 100x784
  err = err +  1/N*sum(sum( (data(:,1:end-1)-dataout).^2 ));
 end
 test_err(epoch)=err/testnumbatches;
 fprintf(1,'Before epoch %d Train squared error: %6.3f Test squared error: %6.3f \t \t \n',epoch,train_err(epoch),test_err(epoch));

%%%%%%%%%%%%%% END OF COMPUTING TEST RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  %%%% DISPLAY FIGURE TOP ROW REAL DATA BOTTOM ROW RECONSTRUCTIONS %%%%%%%%%%%%%%%%%%%%%%%%%
 fprintf(1,'Displaying in figure 2: Top row - real test data, Bottom row -- reconstructions \n');
 output=[];
 for ii=1:15
  output = [output data(ii,1:end-1)' dataout(ii,:)']; % 784x100 ++ 784x1
 end
   if epoch==1
   %close all
   figure('Position',[100,600,1000,200]);
   else
   figure(1)
   end
   mnistdisp(output);
   drawnow;
   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 tt=0;
 for batch = 1:numbatches/10
  fprintf(1,'epoch %d batch %d\n',epoch,batch);

  %%%%%%%%%%% COMBINE 10 MINIBATCHES INTO 1 LARGER MINIBATCH %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  tt=tt+1;
  data=[];
  for kk=1:10
   data=[data
        batchdata(:,:,(tt-1)*10+kk)];
  end

%%%%%%%%%%%%%%% PERFORM CONJUGATE GRADIENT WITH 3 LINESEARCHES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  max_iter=3;
  VV = [w1(:)' w2(:)' w3(:)' w4(:)' w5(:)' w6(:)' w7(:)' w8(:)']';
  Dim = [l1; l2; l3; l4; l5; l6; l7; l8; l9];

  [X, fX] = minimize(VV,'CG_MNIST',max_iter,Dim,data);

  w1 = reshape(X(1:(l1+1)*l2),l1+1,l2);
  xxx = (l1+1)*l2;
  w2 = reshape(X(xxx+1:xxx+(l2+1)*l3),l2+1,l3);
  xxx = xxx+(l2+1)*l3;
  w3 = reshape(X(xxx+1:xxx+(l3+1)*l4),l3+1,l4);
  xxx = xxx+(l3+1)*l4;
  w4 = reshape(X(xxx+1:xxx+(l4+1)*l5),l4+1,l5);
  xxx = xxx+(l4+1)*l5;
  w5 = reshape(X(xxx+1:xxx+(l5+1)*l6),l5+1,l6);
  xxx = xxx+(l5+1)*l6;
  w6 = reshape(X(xxx+1:xxx+(l6+1)*l7),l6+1,l7);
  xxx = xxx+(l6+1)*l7;
  w7 = reshape(X(xxx+1:xxx+(l7+1)*l8),l7+1,l8);
  xxx = xxx+(l7+1)*l8;
  w8 = reshape(X(xxx+1:xxx+(l8+1)*l9),l8+1,l9);

  %%%%%%%%%%%%%%% END OF CONJUGATE GRADIENT WITH 3 LINESEARCHES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%

 end

 save mnist_weights w1 w2 w3 w4 w5 w6 w7 w8
 save mnist_error test_err train_err;

end