simufunc.m

我ANN的大作业

function FunApprox()

%获取训练样本

target_T = zeros(10,1);
unit = 9;
for i = 0:unit
    for j= 0:unit
        index = i*unit+j+1;
        input_P(index,1) = 0.1*i;
        input_P(index,2) = 0.1*j;
        target_T(index) = f(input_P(index,:));
    end
end
input_P
target_T

test_T = zeros(10,1);
unit = 99;
for i = 0:unit
    for j= 0:unit
        index = i*unit+j+1;
        test_P(index,1) = 0+0.01*i;
        test_P(index,2) = 0+0.01*j;
        test_T(index) = f(test_P(index,:));
    end
end

%训练三层神经网络,隐藏层有6个单元
eps = 0.05;
alpha = 0.1;
initU1 = [0.43 0.54 
          0.23 0.54 
          0.54 0.54 
          0.53 0.22 
          0.15 0.98
          0.67 0.23 ];
initV1 = [0.43 0.56 
          0.24 0.5 
          0.54 0.26 
          0.76 0.87 
          0.26 0.38
          0.38 0.48 ];
initb1 = [0.42;0.23;0.31;0.43;0.42;0.34];

initU2 = [0.5 0.55 0.32 0.21 0.23 0.6];
initV2 = [0.4 0.32 0.52 0.36 0.14 0.9];
initb2 = [0.1];

[u1,u2, v1, v2, b1, b2, times] = MLP(eps, initU1, initV1, initb1, initU2, initV2, initb2 , alpha, input_P, target_T)

%验证神经网络的正确性
len = length(test_T);
correct = 0;
for i=1:len
    x0 = test_P(i,:)';
    x1 = sigmoid(u1*(x0.^2) + v1*x0 + b1);
    x2 = sigmoid(u2*(x1.^2) + v2*x1 + b2);
    if(abs(x2-test_T(i))<=eps)
        correct = correct+1;
    end
    test_T(i) = x2;
end
correctRate = correct/len;

len = length(test_P);
actualZ = zeros(10,1);
for i=1:len
    actualZ(i) = f(test_P(i,:));
end

figure
plot3(test_P(:,1), test_P(:,2), actualZ);
legend('f(x,y)');
figure
plot3(test_P(:,1), test_P(:,2), test_T);
legend('Network Result');
figure
plot3(test_P(:,1), test_P(:,2), actualZ, test_P(:,1), test_P(:,2), test_T);
legend('f(x,y)','Network Result');

function y=f(x)
y = (1+sin(6*(x(1)^2+x(2)^2)))/2;