Does patternnet create MLP neural network??

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???

Answer 1

[Greg Heath]

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

[Walter Roberson]

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]

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

Answer 2

[Walter Roberson]

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.

[Greg Heath]

Incorrect.

MLP is a topology.

It can be trained by a number of different algorithms.

Greg

[Samisam]

I Know but my question is about the command According to your answer I can create it using Patternnet can't I ?

[Samisam]

and if I can I would be thankful if you give me the syntax to create nn with 4 input - 2 hidden layers each one with 10 neuron and 1 neuron in out put layer using patternnet thanks any way and witch command in your opinion is better to train my nn with abc algorithm is it patternnet or newff or what ?????

[Greg Heath]

1. ONE hidden layer is sufficient.
2. Use FITNET for curveFITting and regression.
3. Use PATTERNNET for PATTERN-recognition and classification.
4. "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.
5. 10 neurons in a hidden layer is a default that does not have to be specified
6. The basic code for each is given in the help and doc documentation:

 help fitnet
 doc fitnet

and similarly for patternnet.

In addition, I have posted zillions of examples, including tutorials, in both the NEWSGROUP and ANSWERS.

[Samisam]

thanks