Output of Feedforward Network does not match manual calculations

1 次查看（过去 30 天）

显示更早的评论

Guilherme Candido 2019-6-17

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/467380-output-of-feedforward-network-does-not-match-manual-calculations

编辑： Guilherme Candido 2019-6-20

在 MATLAB Online 中打开

Hi,

I have trained a small network with 1 hidden layer with 10 nodes..

iInputsSize=4; % 4 inputs
iOutputsSize=3; % 3 outputs 
sTrainFcn = 'trainlm';  % Function
iHiddenLayerSize = 10; % Hidden Layer Size
net = fitnet(iHiddenLayerSize,sTrainFcn);
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100; 
net.performFcn = 'mse'; 
net.performParam.regularization = 0; 
net.trainparam.epochs = 1000;
net.trainParam.goal = 0;
net.trainParam.mu = 1.0000e-03;
net.trainParam.max_fail = 6;
net.layers{1}.transferFcn = 'tansig';
net.layers{2}.transferFcn = 'purelin';
[net,tr] = train(net,x,t);
%% Test
x=[0.002982739	-0.004732588	0.012710054	0.002787759]
y = net(x); 

...and tried to match its output doing manual calculations. The outputs do not match.

>> x=[0.002982739	-0.004732588	0.012710054	0.002787759]'
x =
    0.0030
   -0.0047
    0.0127
    0.0028
>> y=net(x)
y =
    0.0307
    0.0036
    0.0175