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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
| function [y]=ReLU(x) //RELU FUNCTION
y= max(0,x)
endfunction
function [y]= Softmax(x) //SOFTMAX FUNCTION
ex = exp(x);
y = ex/sum(ex);
endfunction
function [y] = Pool(x) // POOLING FUNCTION
[xrow,xcol,numFilters] = size(x);
y= zeros (xrow/2,xcol/2,numFilters);
for k=1:numFilters
filt = ones(2,2)/(2*2)
image = conv2(x(:,:,k),filt, 'valid');
[irow,icol,numFilters] = size(image);
y(:,:,k) = image (1:2:irow,1:2:icol)
end
endfunction
function [y] = Convo(x,W) // CONVOLUTION FUNCTION
[wrow,wcol,numFilters] = size(W);
[xrow,xcol] = size(x);
yrow = xrow-wrow+1;
ycol = xcol-wcol+1;
y = zeros (yrow,ycol,numFilters);
for k = 1:numFilters
filt = W(:,:,k);
//disp (filt)
filt = filt(wrow:-1:1,wcol:-1:1)
//disp(size(y))
y(:,:,k) = conv2 (x,filt,'valid')
end
end
function images = loadMNISTImages(filename) //MNIST DATASET IMAGES LOADING FUNCTION
fd = mopen(filename, 'r')
magic = mget(1, 'ib', fd) //Be sure to use the right name and path to the file
numImages = mget(1, 'ib', fd)
numRows = mget(1, 'ib', fd)
numCols = mget(1, 'ib', fd)
images = mgeti(numImages*numRows*numCols, 'uc', fd)
mclose(fd)
images = matrix(images, numCols, numRows, numImages)
images = permute(images, [2 1 3])
images = matrix(images,size(images, 1),size(images, 2),size(images, 3))
images = double(images)/255
endfunction
function labels= loadMNISTLabels(filename) // MNIST DATASET LABELS LOADING FUNCTION
fd = mopen (filename,'r')
magic = mget(1, 'ib', fd) //Be sure to use the right name and path to the file
numLabels = mget(1, 'ib', fd)
labels = mget (numLabels,'uc',fd)
mclose(fd)
endfunction
function [W1,W5,Wo]= MNISTConv (W1,W5, Wo, X, D) //TRAINING FUNCTION
alpha = 0.01;
bet = 0.95;
momentum1 = zeros (9,9,20);
momentum5 = zeros (100,2000);
momentumo = zeros (10,100);
N = length(D);
bsize=100; //setting a batch size
blist=1:bsize:(N-bsize+1); //finding all the starting indice of the different batchs
for batch = 1:length(blist)
disp(batch) //keeping the user updated about the number of the current batch
dW1 = zeros(9,9,20);
dW5 = zeros(100,2000);
dWo = zeros(10,100);
begin = blist(batch);
for k= begin:begin+bsize-1 //propagation
x = X(:,:,k);
y1 = Convo(x,W1);
y2 = ReLU (y1);
y3 = Pool (y2);
y4 = matrix (y3,-1,1);
v5 = W5*y4;
y5 = ReLU(v5);
v = Wo*y5;
y = Softmax(v);
d = zeros (10,1); //comparing to expected result
d(sub2ind(size(d),D(k),1))=1;
e=d-y; //calculating error vector
delta =e;
e5= Wo'*delta;
delta5 = (y5 > 0) .*e5; //backpropagation
e4 = W5'*delta5;
e3 = matrix(e4,size(y3));
e2 = zeros(20,20,20);
W3 = ones (20,20,20)/(2*2);
for c = 1:20
e2(:,:,c) = e3(:,:,c).*.ones(2,2)
e2(:,:,c) = e2(:,:,c).*W3(:,:,c);
end
delta2 = (y2 > 0).*e2;
delta1_x = zeros(9,9,20);
rotdelta2= delta2(20:-1:1,20:-1:1,:)
for c=1:20
delta1_x(:,:,c)= conv2(x(:,:), rotdelta2(:,:,c),'valid');
end
dW1 = dW1 + delta1_x; //saving the error vector in an error "batch pool"
dW5 = dW5 + delta5*y4';
dWo = dWo + delta *y5';
end
dW1 = dW1 / bsize; //creating a "batch error vector"
dW5 = dW5 / bsize;
dWo = dWo / bsize;
momentum1 = alpha*dW1 + bet*momentum1; //applying it
W1 = W1+ momentum1;
momentum5 = alpha*dW5 + bet*momentum5;
W5 = W5+ momentum5;
momentumo = alpha*dWo + bet*momentumo;
Wo = Wo+ momentumo;
end
endfunction
//PROGRAM
Images = loadMNISTImages('t10k-images.idx3-ubyte'); //Loading training images
Images = matrix (Images,28,28,-1);
Labels = loadMNISTLabels('t10k-labels.idx1-ubyte') //Loading traininn images labels
Labels (Labels == 0) = 10
W1 = 2*rand(9,9,20)-1 //Randomly asigning wheight to convolution filters, and Neural connection
W5= (2*rand(100,2000)-1) * sqrt(6)/sqrt(360+2000);
Wo = (2*rand(10,100) -1) * sqrt(6)/sqrt(10 + 100);
X = Images(:,:, 1:8000); //Selecting the first 8000 images for training
D = Labels(1:8000);
for epoch =1:3
disp('\\\\\',epoch, '/////') //Training the network by using the training function.
[W1,W5,Wo] = MNISTConv (W1,W5,Wo,X,D); //Inputs are the weigths, Images and Labels,
//Output are the new weights, that we will send again in the next epoch, or test.
end
X = Images(:,:, 8001:10000); //Selecting the last 2000 images we didn't use to use them as a test for the network
D = Labels(8001:10000); //Same for the labels
acc = 0;
N = length(D);
disp('test') //Beginning of the test
for k= 1:N
x = X(:,:,k); //Propagation
y1 = Convo(x,W1);
y2 = ReLU(y1);
y3 = Pool (y2);
y4 = matrix (y3, -1, 1);
v5 = W5*y4;
y5 = ReLU(v5);
v = Wo*y5;
y = Softmax (v);
[m,i ]= max(y); //Calculating accuracy
if i == D(k)
acc = acc+1;
end
end
acc = acc/N;
disp ('accuracy is') //Dysplaying accuracy
disp (acc); |
MNIST ConvNet : Comment charger et formater les données?
Répondre avec citation
Envoyé par Jerome Briot
Partager