Learning SAS? Welcome to the exclusive online community for all SAS learners.

Neural Networks in SAS Enterprise Miner training

Accepted Solution Solved
Reply
Contributor
Posts: 23
Accepted Solution

Neural Networks in SAS Enterprise Miner training

I am using the "Applied Analytics using SAS Enterprise Miner" courseware (AAEM41) and currently looking at Chapter 5 using Neural Networks.  I am using Enterprise Miner on SAS OnDemand, version 14.3.

 

When I go through the PVA97nk sample in the textbook, none of my results ever match what is in the courseware.  This includes number of weights (i., estimates) and other details.  For all the other chapters, my results match the courseware.  This is not a big deal but I am just wondering if the Neural Network algorithms have changed in and are thus causing the mismatch between my results and the manual.  

I am just wondering if I am doing something wrong that I need to correct.


Accepted Solutions
Solution
‎04-02-2018 08:31 AM
SAS Super FREQ
Posts: 9,424

Re: Neural Networks in SAS Enterprise Miner training

Hi:
I understand from the instructors who teach this class that it is entirely possible for you to receive different results than the ones we show in the book or the e-learning class (whichever one you are using).

For example, in the "old" days when it was possible for some customers to have servers on a 32 bit processor or on a 64 bit processor, it was possible for the same EM node to generate different results based on the processor being used. This was explained to me as " there is actually near-hardware layer (something about the instruction layer within the computing core) dependency within EM" -- which might cause different computers to get different results. This is one reason why we control the image that candidates use to take the Predictive Modeler exam, so that they get the correct answer when using EM for the exam.

Or, as you note, the course code you provide was written and all the exercises were run using EM 14.1. If you are using a different version of EM, it is entirely possible for your results to differ.

Our most recent course notes, that we use in our training centers are for Enterprise Miner version 14.2, so unfortunately, I do not have version 14.3 course notes available.

Hope this explains some possible reasons why you might be getting different results.

Cynthia

View solution in original post


All Replies
Solution
‎04-02-2018 08:31 AM
SAS Super FREQ
Posts: 9,424

Re: Neural Networks in SAS Enterprise Miner training

Hi:
I understand from the instructors who teach this class that it is entirely possible for you to receive different results than the ones we show in the book or the e-learning class (whichever one you are using).

For example, in the "old" days when it was possible for some customers to have servers on a 32 bit processor or on a 64 bit processor, it was possible for the same EM node to generate different results based on the processor being used. This was explained to me as " there is actually near-hardware layer (something about the instruction layer within the computing core) dependency within EM" -- which might cause different computers to get different results. This is one reason why we control the image that candidates use to take the Predictive Modeler exam, so that they get the correct answer when using EM for the exam.

Or, as you note, the course code you provide was written and all the exercises were run using EM 14.1. If you are using a different version of EM, it is entirely possible for your results to differ.

Our most recent course notes, that we use in our training centers are for Enterprise Miner version 14.2, so unfortunately, I do not have version 14.3 course notes available.

Hope this explains some possible reasons why you might be getting different results.

Cynthia
Contributor
Posts: 23

Re: Neural Networks in SAS Enterprise Miner training

Posted in reply to Cynthia_sas

Cynthia,

 

Thank you very much.  The response does make sense.  I just wanted to make sure that that I hadn't done something wrong.  Since Neural Networks do not provide much in the way of useful explanations of how they generate a prediction, I had little intermediate results to compare. 

 

Jerry

☑ This topic is solved.

Need further help from the community? Please ask a new question.

Discussion stats
  • 2 replies
  • 342 views
  • 0 likes
  • 2 in conversation