Muster/auf72.m at master · jalobe/Muster · 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
clear all;
close all;
clc;


% Daten aus Dateien einlesen
X_train=dlmread('pendigits-training.txt');
X_test=dlmread('pendigits-testing.txt');
y_train = X_train(:,end);
t_train = zeros(size(y_train,1),10);
for i=1:size(y_train)
t_train(i,y_train(i) + 1) = 1;
end
y_test = X_test(:,end);
X_train = X_train(:,1:16);
X_test = X_test(:,1:16);

% Zentriere Daten
u_train = mean(X_train,1);
u_test = mean(X_test,1);
X_train = X_train - repmat(u_train,size(X_train,1),1);
X_test = X_test - repmat(u_test,size(X_test,1),1);

% Berechne Kovarianzmatrix von X_train
cov_train = cov(X_train);
% Berechne Eigenvektoren mittels SVD
[u,s,v] = svd(cov_train);
% Basistransformationsmatrix
X_haupt = u;

% Transformiere Daten in neue Basis und reduziere auf 14 Dimensionen
X_test = X_test * X_haupt;
X_train = X_train * X_haupt;
X_test = X_test(:,1:14);
X_train = X_train(:,1:14);


% Subsetgröße
s = 100;
% Anzahl innerer Neuronen
n = 15;
% Lernkonstante
gamma = 10/s;
% Fehlerschwelle
epsilon = .01*s;

% random weights
W1_hat = rand(15,n);
W2_hat = rand(n+1,10);

count = 0;
do
	count += 1;
	E = 0;
	dW1_hat = 0;
	dW2_hat = 0;

	for(j=1:s)
		% Input
		i = randi(size(X_train,1));
		[grad1,grad2,error] = backprop(X_train(i,:),W1_hat,W2_hat,t_train(i,:));
		E += error;
		dW1_hat += -gamma * grad1;
		dW2_hat += -gamma * grad2;
	endfor

	Es(count,1) = E;
	W1_hat += dW1_hat;
	W2_hat += dW2_hat;

until(E < epsilon)


plot(Es);

hits = 0;
datasize = size(X_test,1);
for(i=1:datasize)
	class = classify(W1_hat,W2_hat,X_test(i,:));
	if(class == y_test(i))
		hits += 1;
	endif
end
ER = hits/datasize *100