PatternRecogProject/digit_classify.m at master · JoniKet/PatternRecogProject · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
% Digit classify main function made for Pattern recognition course.
% Loads in hyperparameters and neural network weights from NetworkStructure.mat file
% Loads minumum values and maximum values from maxForNormalization and
% minForNormalizations
% IN ORDER TO RUN THIS, digit_classify_parameters and usedFunctions folders
% have to be added to MATLAB path!!!

function C = digit_classify(testdata)

    parameters = load("NetworkStructure.mat");
    normMin = load("minForNormalization.mat");
    normMax = load("maxForNormalization.mat");


    % ------------------------------------------------------------
    % Given timeseries preprocessing

    smoothed = expsmooth(testdata(:,1:2),500,10); % Exponential smoothing, already dropping Z axis (height)

    [eigenvectors,eigenvalues] = eig(cov(smoothed));
    eigenvector11 = eigenvectors(1,1);
    eigenvector12 = eigenvectors(2,1);
    eigenvector21 = eigenvectors(1,2);
    eigenvector22 = eigenvectors(2,2);
    eigenvalue1 = eigenvalues(1,1);
    eigenvalue2 = eigenvalues(2,2);

    centered = center(smoothed); % Centering

    rescaled = minmaxnorm(centered); % Digit normalization

    %------------------------------------------------------------
    % Feature extraction
    pixelWidth = 12; % hyperparameter of the solution that could increase accuracy
    pixels = zeros(pixelWidth,pixelWidth);
        [pixels,startX,startY,afterstartX,afterstartY,...
            beforeendX,beforeendY,endX,endY,startDirX,...
            startDirY,afterstartdirX,afterstartdirY,...
            beforeenddirX,beforeenddirY] = time2matrix2(rescaled,pixelWidth);

        dataWrangled(1,1:pixelWidth*pixelWidth) = reshape(pixels,1,[]);

        k = 1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = startX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = startY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = afterstartX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = afterstartY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = beforeendX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = beforeendY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = endX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = endY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = startDirX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = startDirY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = afterstartdirX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = afterstartdirY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = beforeenddirX;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = beforeenddirY;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvector11;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvector12;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvector21;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvector22;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvalue1;
        k = k+1;
        dataWrangled(1,pixelWidth*pixelWidth+k) = eigenvalue2;


    [temp,features] = size(dataWrangled);

    %temporarily adding min and max values of features in order to do
    %minmaxnormalization

    dataWrangled(2,:) = normMin(1).minForScaling;
    dataWrangled(3,:) = normMax(1).maxForScaling;

    %normalize the feature columns
    dataWrangled(:,pixelWidth*pixelWidth:features) = minmaxnorm(dataWrangled(:,pixelWidth*pixelWidth:features));
    dataWrangled = dataWrangled(1,:); % Removing the temporary min max rows


    %---------------------------------------------------------------------------
    % Neural Network classifier
    dataWrangled = dataWrangled';
    NNconfig = parameters(1).NNconfigStored;
    wHidden = parameters(1).wHiddenStored;
    wOutput = parameters(1).wOutputStored;
    actFunc = 'sigma';
    extendedInput = [dataWrangled; ones(1, size(dataWrangled, 2))];
    N = size(dataWrangled, 2); % number of data samples

    for i = 1:length(NNconfig) % for each hidden layer

      if i == 1
          yHidden{i} = forwardpropagate(extendedInput,wHidden{i},actFunc,N);
      else
          yHidden{i} = forwardpropagate(yHidden{i-1},wHidden{i},actFunc,N);
      end
    end

    vOutput = wOutput' * yHidden{end}; % output layer weighted inputs
    yOutput = vOutput; % output layer (linear) output

    [~, testClass] = max(yOutput, [], 1);
    C = testClass-1;

end