topic Re: SAS Viya VDMML Solution with By Variable in SAS Data Science

SAS Viya VDMML Solution with By Variable

C_Golen — Wed, 26 Apr 2023 07:52:36 GMT

Hi to all!

I have a data set on SAS Viya version 3.5. I set up an ml model to predict how much the specific product will be sold for each store in my data set. However, while setting up this model, I want to give "By Variable" just like in the forecast studio. In other words, I want to model on a store-item basis with ML solution.

Previously in Enterprise Miner we could do this with start end group. How can we do it in Viya? Also, how to do K-fold cross validation in SAS Viya?

Ref: https://communities.sas.com/t5/SAS-Communities-Library/How-to-calculate-cross-validation-error-using-the-Start-and-End/ta-p/30724

Thanks in advance!

Re: SAS Viya VDMML Solution with By Variable

sbxkoenk — Wed, 26 Apr 2023 13:44:04 GMT

Hello,

I am afraid that this "Start Group - End Group" functionality from Enterprise Miner has disappeared in VDMML (when using Model Studio front-end app). I guess it's on the to-do list of Model Studio R&D.

With regard to Cross-Validation (CV) :
Are you wondering how to perform cross-validation inside a Model Studio VDMML pipeline?

In Enterprise Miner I was also using "Start Group - End Group" functionality to achieve that.

Haven't tried to do CV in Model Studio yet.

Outside Model Studio , ... in SAS Studio ... , CV is easy to program.
Let me know if you want an easy program to create the k folds.
See also this :

crossValidateML action in SAS Viya 3.5
https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/casactml/casactml_mltools_details02.htm

Here's an example of its usage :

SAS Help Center: Example 20.1 Cross Validation of a Forest Model

Koen

Re: SAS Viya VDMML Solution with By Variable

C_Golen — Wed, 26 Apr 2023 21:34:21 GMT

Hi Koen,

Many thanks for your interest.

Sorry to hear that 😕
My main goal is use it in the front - end. So, it would be great if I can manage to add this option to our pipeline.

Re: SAS Viya VDMML Solution with By Variable

sbxkoenk — Thu, 27 Apr 2023 15:10:44 GMT

Hello,

What exactly do you want the cross-validation for? :

- to honestly estimate the generalization error of the model (detect over-fitting)

- to find the best hyper-parameters

- to ...

What type of model do you want to fit?

Cross validation is offered within gradient boosting as part of the VDMML.
Both gradient boosting and forest models in SAS Model Studio allow for hyper-parameter optimization and cross validation to optimize model properties.

Koen