CNN-MNIST-MATLAB/CNN.m at master · AshishMehtaIO/CNN-MNIST-MATLAB · 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
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
%% Code to import training data from MNIST dataset

clear all;

train_images = reshape((loadMNISTImages('train-images.idx3-ubyte')),[28,28,1,60000]);
train_labels = ((loadMNISTLabels('train-labels.idx1-ubyte')));


test_images = reshape((loadMNISTImages('t10k-images.idx3-ubyte')),[28,28,1,10000]);
test_labels = ((loadMNISTLabels('t10k-labels.idx1-ubyte')));


%% Code to generate 10 fold train data

train_images_fold=zeros(28,28,1,6000,10);
train_labels_fold=zeros(6000,1,10);

for i=(1:10)
train_images_fold(:,:,:,:,i)=train_images(:,:,:,(6000*(i-1))+1:6000*i);
train_labels_fold(:,1,i)=train_labels((6000*(i-1))+1:6000*i,1);
end

train_images_data=zeros(28,28,1,54000,10);
train_labels_data=zeros(54000,1,10);
cross_valid_images=zeros(28,28,1,6000,10);
cross_valid_labels=zeros(6000,1,10);

for (i=1:10)
    cross_valid_images(:,:,:,:,i)=train_images_fold(:,:,:,:,i);
    cross_valid_labels(:,:,i)=train_labels_fold(:,:,i);
    k=1;
    for j=(1:10)
        if (j~=i)
           train_images_data (:,:,:,(6000*(k-1))+1:6000*k,i)=train_images_fold(:,:,:,:,j);
           train_labels_data ((6000*(k-1))+1:6000*k,1,i)=train_labels_fold(:,:,j);
           k=k+1;
        end
    end
end


%% Code to display first 100 training images
figure(1)
for i=1:100
        subplot(10,10,i);
        imshow(train_images(:,:,:,i));
end

%% Code to generate the ANN layers

inputlayer = imageInputLayer([28 28 1],'DataAugmentation','none',...
    'Normalization','none','Name','input');

%%volume 28*28*1
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%                Layer 1
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

convlayer1 = convolution2dLayer(4,32,'Stride',1,'Padding',0, ...
    'BiasLearnRateFactor',2,'NumChannels',1,...
    'WeightLearnRateFactor',2, 'WeightL2Factor',1,...
    'BiasL2Factor',1,'Name','conv1');

%%volume 25*25*32

convlayer1.Weights = randn([4 4 1 32])*0.1;
convlayer1.Bias = randn([1 1 32])*0.1;

relulayer1 = reluLayer('Name','relu1');

localnormlayer1 = crossChannelNormalizationLayer(3,'Name',...
    'localnorm1','Alpha',0.0001,'Beta',0.75,'K',2);

maxpoollayer1 = maxPooling2dLayer(3,'Stride',3,'Name','maxpool1','Padding',1);

%volume 9*9*32

droplayer1 = dropoutLayer(0.35);


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%                  Layer 2
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

convlayer2 = convolution2dLayer(3,16,'Stride',1, 'Padding',0,...
    'BiasLearnRateFactor',1,'NumChannels',32,...
    'WeightLearnRateFactor',1, 'WeightL2Factor',1,...
    'BiasL2Factor',1,'Name','conv2');

%7*7*16

convlayer2.Weights = randn([3 3 32 16])*0.0001;
convlayer2.Bias = randn([1 1 16])*0.00001;

relulayer2 = reluLayer('Name','relu2');

localnormlayer2 = crossChannelNormalizationLayer(3,'Name',...
    'localnorm2','Alpha',0.0001,'Beta',0.75,'K',2);

%maxpoollayer2 = maxPooling2dLayer(2,'Stride',2,'Name','maxpool1','Padding',1);

droplayer2 = dropoutLayer(0.25);


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%                  Output Layers
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

fullconnectlayer = fullyConnectedLayer(10,'WeightLearnRateFactor',1,...
    'BiasLearnRateFactor',1,'WeightL2Factor',1,'BiasL2Factor',1,...
    'Name','fullconnect1');
fullconnectlayer.Weights = randn([10 784])*0.0001;
fullconnectlayer.Bias = randn([10 1])*0.0001+1;

smlayer = softmaxLayer('Name','sml1');

coutputlayer = classificationLayer('Name','coutput');


%% Code to define training parameters

options = trainingOptions('sgdm',...
      'LearnRateSchedule','piecewise',...
      'LearnRateDropFactor',0.75,...
      'LearnRateDropPeriod',1,'L2Regularization',0.0001,...
      'MaxEpochs',16,'Momentum',0.9,'Shuffle','once',...
      'MiniBatchSize',15,'Verbose',1,...
      'CheckpointPath','E:\ccfuser3\checkpoints','InitialLearnRate',0.043);

 %% Code to make the network
    layers =[inputlayer, convlayer1, relulayer1,localnormlayer1, ...
       maxpoollayer1, droplayer1,...
       convlayer2, relulayer2, localnormlayer2,droplayer2,...
       fullconnectlayer, smlayer, coutputlayer];


   %% Train the ANN
   train_im=zeros(28,28,1,54000);
   train_lb=categorical(zeros(54000,1));
   cross_valid_im=zeros(28,28,1,6000);
   cross_valid_lb=zeros(6000,1);
   for (i=1:10)
       train_im=train_images_data(:,:,:,:,i);
       train_lb=categorical(train_labels_data(:,:,i));
       cross_valid_im=cross_valid_images(:,:,:,:,i);
       cross_valid_lb=categorical(cross_valid_labels(:,:,i));
trainedNet(i) = trainNetwork(train_im,train_lb,layers,options);

[Ypred,scores] = classify(trainedNet(i),cross_valid_im);
score(i) = sum((Ypred==cross_valid_lb))/numel(cross_valid_lb)
   end

   [max,index]=max(score);
   best_model=trainedNet(index);