topic Re: Scoring using model from open source node (Python) - VDMML in SAS Data Science

Scoring using model from open source node (Python) - VDMML

andreas_zaras — Tue, 13 Jul 2021 19:59:37 GMT

Hello,

I have developped the pipeline presented in the attached screenshot. A decision tree is developped in Python - Open SOurce Node. Then i add a Score node as a child and i attach a score data set. The node runs and a data set is output but on;y with the input variables (the probabilities and decision variables do not appear e.g. EM_Classification, EM_Event_Probability).

In order to score i need to write code in the python open source node or i am doing something wrong? The score node facility cannot be used in the case of an open source model?

Thanks in advance,

Andreas

Re: Scoring using model from open source node (Python) - VDMML

HarrySnart — Thu, 15 Jul 2021 15:00:48 GMT

Hi @andreas_zaras , are you using SAS Viya 3.5 here?

To use the Score Data Node I believe that you need to have an associated score code file. In SAS Viya 3.5 the Open Code node generates a .py file rather than a scoring file. To score the model like you are doing here you could add an open code node afterwards to score the model.

You should be able to view performance of the open model against your analytic partitions in the Pipeline Comparison tab already, as you can feed in the same analytic partition you define when you create the project.

To get a scoring file generated for your Python model you can add it to model manager using the register_model utility available in the sasctl Python library. There is an example here on Github: https://github.com/sassoftware/python-sasctl

In SAS Viya 4 you can add a train.py and score.py script to the Open Code node directly in Model Studio which streamlines the process a bit.

I hope this helps

Thanks,

Harry

Re: Scoring using model from open source node (Python) - VDMML

andreas_zaras — Mon, 19 Jul 2021 20:13:04 GMT

Hi!

Thanks for your answer! I am usingg Viya 3.5. I am able to see performance measures through the model comparison tab.

Thanks again,

Andreas