BookmarkSubscribeRSS Feed
PaalNavestad
Pyrite | Level 9

Hi, i have a dataset which consist of cells in space. For each cell I have measurements at 16 different times. We are really uncertain around what are he best times and cells to include in model building. My idea was to create similar models where we based on rules wanted to include different cells and times in the model training. However I can not find a way to create different subsets from the main Data set in Model Studio. The filter node does only filter based on outliers/distribution it seems. I knew that in Enteprise Miner I could run and compare models based on different subsets. 

 

To exemplify I have the group identifier A,B on the data. I would really like to build a model for only A, only B and A and B combined. Then I would like to compare the models either in one pipeline or in different pipelines inside Model Studio. I can't find a way tp do this.

 

Can anybody help?

Thanks 

2 REPLIES 2
tom_evans_79
Fluorite | Level 6

Hi Paal,

 

Old post, I know, but I'm working on something similar and might have a partial answer to your issue.

 

I added a SAS Code node after the data node. I left the "Training Code" section blank. In the "Scoring Code" section, I added "if seg_flg=1;", where seg_flg=1 is the logic I'm using to define my first subset. Then I added modelling nodes after the SAS Code node. I repeated this for each subset.

 

In your situation, this may work if you want to evaluate the models based on the specific subset only, rather than the whole population. On the other hand, if, like me, you want to stack the scored subsets and evaluate them together, then there's something missing.

 

If you prefer to use filters, this might be helpful: https://github.com/sassoftware/sas-viya-dmml-pipelines/tree/master/sas_code_node/subset_data

 

/Tom.

PaalNavestad
Pyrite | Level 9

Thanks Tom, This might work in the use case I was working. I'll keep it in mind. In the original as I found no fast solution I did it all with code. That works but loose some o f the help in documentation etc.

 

Ready to join fellow brilliant minds for the SAS Hackathon?

Build your skills. Make connections. Enjoy creative freedom. Maybe change the world. Registration is now open through August 30th. Visit the SAS Hackathon homepage.

Register today!
How to choose a machine learning algorithm

Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 2 replies
  • 941 views
  • 1 like
  • 2 in conversation