Builder NE 2010 a problem with neural network toolbox

Question

ali 2011-7-15

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/11679-builder-ne-2010-a-problem-with-neural-network-toolbox

Hi, I'm using 2010a, trying to read a saved neural network and then simulate it using an input which is passed from a C# windows form. I created two projects using deploytool. the first one reads the .mat file and returns the network1 object containing the net;

function answer = readnn(path) load(path); answer = network1; end

second one simulates it using sim command:

function answer = simnn(net, input) dummy = network(net); answer = sim(dummy,input); end

I reference both projects in VS2010. the read part works just fine and I can see the net in debugger(although it warnings about a network object being converted to a struct). But it does not simulate.

"MWMCR EvaluateFunction error ... Error using ==> feval undefined function or method 'dotprod' for input argument of type struct"

any thoughts? I created and trained nn in 2010a. also tested ther application in VS2008 in .NET 2.0 3.5 and 4.0 platforms. same problem persists.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Friedrich 2011-7-15

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/11679-builder-ne-2010-a-problem-with-neural-network-toolbox#answer_15938

在 MATLAB Online 中打开

Hi,

I think I have seen this before. When compiling the class constructor is not compiled and during runtime the class in unknown and because of this it gets converted to a struct. (could you post the full warning here?). I think the solution was to add the constructor as pragma:

%#function constructor_name

Unfortunately I have forgotten which the right constructor is. Maybe the warning can help you here.

I think there other problems with compiling a pre trained network. If I can remember correctly a warning like this indicates it:

Undefined function or method 'initzero' for input arguments of type 'char'.

The solution for this is to get in contact with the technical Support. I think they can provide a patch for this.