MATLAB Answers

# Timedelaynet output calculation principle

조회 수: 4(최근 30일)
Tingting Zhang 8 Sep 2019
댓글: Tingting Zhang 10 Sep 2019
I have built a focused time-delay NN (two delays, 2 layers, hidden layer size of 2) and changed all the inputweights and layerweights to be zero (no bias is used). I would expect all zero output, but why the code give me nonzero values. Is there anybody can give suggestions on what's wrong? The simple code is shown as following:
clear;
numtaps = 2;
hiddenLayerSize = 2;
dyn_net = timedelaynet(1:numtaps,hiddenLayerSize);
dyn_net.trainFcn = 'trainlm';
dyn_net.layers{1}.transferFcn = 'tansig';
dyn_net.initFcn = 'initlay';
dyn_net.performFcn = 'mse';
dyn_net.biasConnect = [0;0];
dyn_net.trainParam.epochs = 1;
inputs = 0:0.1:1;
targets= 0:0.1:1;
inputs_1c = num2cell(inputs,1);
targets_1c = num2cell(targets,1);
[p,Pi,Ai,t] = preparets(dyn_net,inputs_1c,targets_1c);
[dyn_net,dyn_tr] = train(dyn_net,p,t,Pi,Ai);
dyn_net.IW{1,1} =[0 0;0 0];
dyn_net.LW{2,1} =[0 0];
output = dyn_net(p,Pi);
outputs1 = cell2mat(output); % I should expect zero, but the outputs are all 0.6.
dyn_net.LW{2,1}*tansig(dyn_net.IW{1,1}*(inputs(6:7)).') % This calculation gives me zero

댓글을 달려면 로그인하십시오.

### 답변(1개)

Greg Heath 9 Sep 2019
You did not include tHe 2 biases.
Hope this helps.
Greg
THANK YOU FOR FORMALLY ACCEPTING MY ANSWER
##### 댓글 수: 1표시숨기기 없음
Tingting Zhang 10 Sep 2019
Hi Greg,
Thanks for your answer. The configure of the NN is shown as following. I intend not to include the bias just want to check the feedforward output calculation. I would expect that to be:
dyn_net.LW{2,1}*tansig(dyn_net.IW{1,1}*(inputs(k-1:-1:k-2)).').
But it is not the same as what I expect. So I manually change the inputweights and layerweights to be zero after training (before training gives me an error), and found nonzero output, that's weird.
I have also tried to include the bias as you suggested, the problem is still there.

댓글을 달려면 로그인하십시오.

### Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!