Does patternnet create MLP neural network??

Question

Samisam 2018-1-3

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/375425-does-patternnet-create-mlp-neural-network

编辑： Walter Roberson 2018-1-5

I want to create a NN to classify Iris Data set with a specific algorithm like (ABC) this NN should be MLP-NN I see a lot of questions and their answer but I can't really consider if patternnet creates MLP-nn or not can any body ensure me???

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

Greg Heath 2018-1-4

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/375425-does-patternnet-create-mlp-neural-network#answer_298683

在 MATLAB Online 中打开

Yes. The configuration generated by the call to patternnet is a

Multi-layer-perceptron

https://en.wikipedia.org/wiki/Perceptron

Thank you for formally accepting my answer

Greg

2 个评论
显示无隐藏无

Walter Roberson 2018-1-4

The documentation at https://www.mathworks.com/help/nnet/ref/perceptron.html specifically says that patternnet does nonlinear separation and that perceptrons never do.

Greg Heath 2018-1-5

编辑：Greg Heath 2018-1-5

Quite a bit of confusion occurs because there is a lack of understanding concerning the term "perceptron" because

The single term perceptron DOES NOT IMPLY HIDDEN LAYERS.

The acronym MLP implies a multilayer perceptron.

Therefore one has to read carefully to be sure which one the referral concerns.

The default configurations of fitnet and patternet have a single hidden layer and are, therefore, MLPs. HOWEVER, the default of 10 can be overwritten to 0. Then the configuration becomes a perceptron.

Hope this helps. (Yeah, I know!)

Greg

请先登录，再进行评论。

Answer 2

Walter Roberson 2018-1-3

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/375425-does-patternnet-create-mlp-neural-network#answer_298579

No, patternnet does not use MLP.

https://www.mathworks.com/help/nnet/ref/perceptron.html

and see the File Exchange for a number of MLP contributions.

5 个评论
显示 3更早的评论隐藏 3更早的评论

Greg Heath 2018-1-5

编辑：Walter Roberson 2018-1-5

在 MATLAB Online 中打开

ONE hidden layer is sufficient.
Use FITNET for curveFITting and regression.
Use PATTERNNET for PATTERN-recognition and classification.
"F"eed"F"orward net new"ff" and special cases new"fit" for curve"FIT"ting and new"pr" for "P"attern "R"ecognition are obsolete.
10 neurons in a hidden layer is a default that does not have to be specified
The basic code for each is given in the help and doc documentation: