Neural Network Results

[Khokhaz]

I have a simple Question 

How to know how many Neural Network Layers and how Many Neural Network Units were used in the model from the results ? 

Thanks  & Regards, 

[Cynthia_sas]

Hi:

  If you are working in one of the classes for the SAS Academy for Data Science, there are 18 classes in the entire program. So, in order to ask the instructors your question, I need to know the following:

1) the name of the class you're asking about

2) the chapter or lesson you're working in

3) whether you're doing a demo or exercise and which one (provide the title of the demo of exercise)

4) which output you're asking about (if there's more than one in the demo or exercise).

  Thanks, in advance, for providing more context for your question.

Cynthia

[Khokhaz]

ّHi Cynthia ,

I am studying Lesson 6 : Neural Networks Using SAS Enterprise Minor under the Module 3: Predictive Modeling.

My question is a general question as I am trying to understand the prediction estimate formula and map it to the results in the output window for the neural network node.

If the variables in the parameter estimates in the neural network node contains H1 only meaning the model have only 1 hidden unit and do we know how many layers the model produced ?





Thanks & Regards,

[Cynthia_sas]

Hi, Thanks for the additional informatioin. I'll pass your question to the instructors who teach this course.

Cynthia

[RobertBlanchard · Accepted solution]

Hi,

The neural network node in SAS enterprise miner allows for at most, only one hidden layer with up to 64 hidden units.  That said, you can use PROC NEURAL to build a neural network with as many hidden layers and hidden units as you would like. The HPNEURAL node allows for more than one hidden layers. 

The "H##" (for example H11) represent the hidden units. The first numeric value represents the hidden layer count and the second represents the hidden unit count.  Therefore, H13 is the third hidden unit, located in the first hidden layer.

There are several ways to surface the weights of the final model.  My favorite method is:
1) From the results window, click on VIEW up at the top
2) Then click MODEL.

3) Then click WEIGHTS FINAL.  This will surface a graph containing the final model weights.  But I prefer to see the actual numeric values. 

4) So with the graph selected, click the TABLE shortcut button (next to the print shortcut button).  The table containing all of the final model weights will appear.

A second method for viewing the final model weights would be to view the score code.  To do this:

1) click VIEW.

2) Click SCORING.

3) Click SAS CODE.

All the best,

  Robert

[Khokhaz]

Thank you for the comprehensive answer.

Regards,