topic Re: Cross Validation type in Enterprise Miner in SAS Data Science

Cross Validation type in Enterprise Miner

husseinmazaar — Fri, 26 Sep 2014 12:07:07 GMT

Dear members,

I need to know how to determine the type of cross validation in neural networks node in enterprise miner?

thanks

Re: Cross Validation type in Enterprise Miner

mohamed_zaki — Fri, 26 Sep 2014 13:23:12 GMT

What is your version of EM ?

I have found other people to refer to the lack of the cross validation in EM with neural network, so they repeat the test several time with different data partition seeds and taking the average.

Check this:

http://www.kms.pwr.wroc.pl/staff/trawinski/wp-content/uploads/2013/09/2009-Comparative-Analysis-of-Neural-Network-Models1.pdf

Hope other to tell us if that possible.

Re: Cross Validation type in Enterprise Miner

husseinmazaar — Sun, 28 Sep 2014 08:17:18 GMT

The version of Enterprise Miner is 6.2. Any way around ?

Thanks

Re: Cross Validation type in Enterprise Miner

husseinmazaar — Mon, 29 Sep 2014 13:48:25 GMT

Any Idea?

Re: Cross Validation type in Enterprise Miner

M_Maldonado — Mon, 29 Sep 2014 14:46:18 GMT

Hi Hussein,

What do you mean by "type of cross-validation" in neural networks?

The Start/End group nodes support cross validation for any model. For a quick example, take a look at page 7 of this paper (http://support.sas.com/resources/papers/proceedings10/123-2010.pdf)...

I can't remember top of my head if you have that option in EM 6.2. If your EM version does not have cross validation option for the Start groups, I suggest go back to the basics. When I read the wikipedia definition of k-fold validation, it sounds like something easy to do in EM. Just an idea, use several Sample nodes, add the same model node to all of them, and use an Ensemble node to combine the posterior predicted probabilities. I have not ran this, but that would be my first try.

Please let me know if this worked OK.

Thanks!

Miguel