Why doesnt the neural network just give me the fitting function as an output?

Question

Ali Almakhmari 2023-8-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2012097-why-doesnt-the-neural-network-just-give-me-the-fitting-function-as-an-output

回答： Udit06 2023-8-29

When I am training a neural network to fit a bunch of data. Why doesn't the Neural Network just give me the function (equation(s)) it developed by the fitting at the very end? Is there a way to do this?

ANN=fitnet([50],'trainbr');
ANN=train(ANN,Input,Output);

2 个评论
显示无隐藏无

Walter Roberson 2023-8-24

The function-like representation of the trained network would involve four 50 x 50 matrices. Most people would not find that readable.

Ali Almakhmari 2023-8-24

Why is it four though? Thats very interesting, I still want to see it. How do you think I can get access to them?

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Udit06 2023-8-29

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2012097-why-doesnt-the-neural-network-just-give-me-the-fitting-function-as-an-output#answer_1295751

在 MATLAB Online 中打开

Hi Ali,

I understand that you want to obtain a function-like representation of the neural network that you have trained. To get a function-like representation of the trained neural network, you can try accessing the weight assigned to each neuron after training and the activation function of each layer of the neural network. You can use the “getwb” function (https://www.mathworks.com/help/deeplearning/ref/getwb.html) to access the weight of each neuron. To access the activation function of each neuron you can use the following.

activationFunction = ANN.layers{layer_num}.transferFcn;

You can refer to the following MathWorks documentation link to understand more about “transferFcn” property: https://www.mathworks.com/help/deeplearning/ug/neural-network-subobject-properties.html#bss4hk6-99

Thus, combining the information obtained, you can get a better understanding of how a function-like representation of a trained neural network looks like.

I hope this helps.