topic Grid search to optimize parameters? in SAS Data Science

Grid search to optimize parameters?

IvanGV — Wed, 10 Jun 2015 15:55:24 GMT

Hi, how are you?

I've been starting to use SAS Enterprise Miner 7.1 H1 and I was wondering if there's any efficent way of programming arbitrary parameters' searches.

For example:

- When tuning K for K-NN, I manually created 10 nodes with 10 different K values.

- When tuning Hidden Units for a Neural Network, I had to do the same thing.

When you want to get greedy with your search space, programming it this way becomes quite tedious.

Is there a way to use a single encapsulating node and easily set different parameters combinations to be tested, instead of having to manually set different model nodes?

Thanks a lot!

Best regards,

Ivan.

Re: Grid search to optimize parameters?

M_Maldonado — Wed, 10 Jun 2015 17:56:53 GMT

Hi Ivan,

This is supported for neural networks using the AutoNeural node.

For other models the only alternative that comes to mind is to use macro programming on base SAS to do multiple calls of a procedure with different parameters.

I hope this helps,

Miguel

Re: Grid search to optimize parameters?

IvanGV — Wed, 10 Jun 2015 18:31:57 GMT

Thank you Miguel, I'm already using AutoNeural so I'll look into macro programming.

If anyone has any code samples to use as reference, I would appreciate those very much, thank you.

Ivan.

Re: Grid search to optimize parameters?

camontanezp — Fri, 22 Sep 2017 22:44:42 GMT

Hi Miguel and Ivan,

I also wonder if there is an example code and if you could guide me to where to write the macro, is it done by adding a SAS code node to the diagram? Thanks.

Re: Grid search to optimize parameters?

AnnaBrown — Mon, 25 Sep 2017 15:28:26 GMT

Hi camontanezp,

Welcome to the SAS Data Mining and Machine Learning Community, thanks for your question! Since it's been quite a while since Miguel and Ivan's discussion, I recommend opening a New Message with your question and reference this thread in that post.

Best,

Anna