Windowing operators (Core vs Value Series Extension)

Noel

Hi@tftemme, [@varunm1,@hughesfleming68]

Hope all is well. Recall the issue I had with the windowing operator misbehaving in 9.4? While that gets sorted out, I switched to the older version of the operator from the Value Series extension and have noticed some strange behavior...

When I have windowed data in the past (with the core operator) that does not have values for the label (e.g. forecasting future periods that aren't able to be labeled... orange box below), the horizon labels would also be blank/null, which makes sense.

In the same situation, the older extension seems to insert values for the horizon label where there was a blank before (blue box below):



Is this a known issue? Does it influence the predictions?

I'm concerned that this operator may have corrupted my process...

Thanks,
Noel

tftemme

Hi@Noel,

You are right. It seems that the Windowing operator of the old value series extension does have a bug with nominal label attributes with missing values. Missing values are just replaced by one value (maybe the last in the order of the mapping of the nominal attribute) of the attribute. So this is definitely problematic. I would recommend to use the new core operator. I posted a fix to your process in the other thread (exclude the indices attribute from the time series attribute selection). This fix the error you are facing there. Also with 9.5.0, the bug itself is fixed.

Best regards,
Fabian

Noel

Hi@tftemme,

Thank you for your help. Much appreciated!