About WendyCzika

WendyCzika · ‎02-26-2018

By default, your segment ID variable from the Cluster node has role of "SEGMENT", and in the Start Groups node, you just need to open the Variables editor, and set the Grouping Role for that variable (_SEGMENT_) to Stratification.

WendyCzika · ‎02-23-2018

You can try the File Import node (from the Sample tab) to read in your data for an Enterprise Miner flow.

WendyCzika · ‎02-19-2018

The only time a node should re-run is if a property in a node before it has changed, or if Rerun is set to Yes in the Input Data node. You might see the green ring around the node briefly when the flow is run, but that doesn't necessarily mean the node is re-running. You can see from the Last Run Time property for the node when it was actually run last.

WendyCzika · ‎02-06-2018

In response to sathya66's question above from a while ago, in case it helps someone else: Q: I have rare Linear targets in my data.( 3000000 observations ,1800 observations are in linear and remaining observations are zeros) How can I oversample the data for interval target ( I am using Linear Regression) or please can you suggest me the procedure to build a model. A: To model data like this, if you create another variable that is an indicator of whether your interval target is > 0 or =0, then set that as a binary target in addition to your interval target, you can use the TwoStage node with the Filter property set to Non-Events.

WendyCzika · ‎02-01-2018

It is on the "HPDM" (high-performance data mining) tab, second node from right, see screenshot.

WendyCzika · ‎01-29-2018

It depends on the structure of your network. With a Parent-child or Markov blanket structure, the inputs can be parents of the target. This paper has all the details and equations of how the predictions for a target are obtained from a Bayesian network: https://support.sas.com/resources/papers/proceedings17/SAS0474-2017.pdf

WendyCzika · ‎01-24-2018

Can you provide more specifics about what node(s) you are using for modeling?

WendyCzika · ‎01-19-2018

It's the data you are scoring that I'm wondering if it has differences: SCORING.DATA_TO_SCORE

WendyCzika · ‎01-19-2018

My guess is that MODEL_SPEC was a character variable in your training data that was used to create the model and score code, and it is numeric in the data you are scoring. It would need to be the same type in both data sets.

WendyCzika · ‎01-09-2018

The log is truncated for me, even when I download it. When you view in Results, can you see the whole log with an error somewhere (usually helps to search on "ERROR:" with the colon) ? If it is truncated there too, you can go to Options>Preferences in EM and change Log/Output Line Numbers to Show All.

WendyCzika · ‎01-08-2018

Yes, that is what's used in PROC NEURAL as well when you have a binary target.

WendyCzika · ‎01-02-2018

When you are viewing the Sample Statistics of the imported or exported data, the type "CLASS" and "VAR" is solely based on whether the variable is character or numeric. But the level of the variable (Nominal or Interval e.g.) if you have used the Advanced Advisor in the Input Data node, is determined by the number of distinct levels of the variable. So by default, if a variable has fewer than 20 levels even if it is numeric, its level will be Nominal and it will be treated as a class variable in the nodes.

WendyCzika · ‎01-02-2018

The HP Forest node can't be used with the Ensemble node since it doesn't create DATA step score code. Try removing that connection and see if that fixes the error.

WendyCzika · ‎12-20-2017

To be able to force particular splits, you would have to use the Interactive Decision Tree Application in the Decision Tree node in EM. Only automated splitting is available in the HP Tree node / PROC HPSPLIT.

WendyCzika · ‎12-19-2017

When you have the project selected in the project panel, click on the "..." for Project Macro Variables, see screenshot.

Online Status	Offline
Date Last Visited	a week ago