Incorrect Neural Network output calculation through weights! Help!

3 次查看（过去 30 天）
Nikos Vasileiadis
Nikos Vasileiadis 2017-10-23
评论： Nikos Vasileiadis 2017-10-23
Hi guys, I confront a problem calculating Neural Network Outputs manually through weights.
The NN has the below specifications:
Input Size 7
1 Hidden Layer of Size 10
Output Size 10
I trained the network train(network,Inputs,Targets) with trainFcn = 'trainlm' and network.layers{1}.transferFcn = 'satlins'. Others defaults.
After training I extracted the weights with the commands below:
b1 = cell2mat(network.b(1)); %Table size 10x1
IW = cell2mat(network.IW); %Table size 10x7
b2 = cell2mat(network.b(2)); %Table size 10x1
LW = cell2mat(network.LW); %Table size 10x10
but when I tried for an input X=[1,1,1,1,1,1,1]':
out1 = purelin( LW * (satlins(IW * X + b1)) + b2 );
out2=network(X);
I took different results out1~=out2.
I have searched similar problems but I cannot adapt them to my problem. Could you help me? Thanks.

请先登录，再回答此问题。

采纳的回答

Greg Heath
Greg Heath 2017-10-23
You did not take into account the default mapminmax normalization of inputs and outputs.
Hope this helps.
Thank you for formally accepting my answer
Greg
  1 个评论
Nikos Vasileiadis
Nikos Vasileiadis 2017-10-23
Thank you Greg! Could you be more specific on how I can modify my calculation to have the right result? I tried to do something like this:
out1 = mapminmax (purelin( LW * (satlins(IW * mapminmax (X) + b1)) + b2 ));
but it didn't work.

请先登录，再进行评论。

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by