Time Series / Moving Average

KNIME Analytics Platform
1

Dear all,

I am just setting up a workflow with the Moving Average node and I experienced that the “time” is shifted by (n-1)/2 rows.
In other words: the moving average is stored with the time stamp from the last value and not with a time stamp of the ((n+1)/2)th value. Therefore the averaged values are (time-wise) shifted compared to the original ones. Is there a possibility to include the discussed possibility as well?

Thank you very much for considering.

Jürgen

2

Dear Jürgen,

you are correct - the current moving average is computed for the last time point of an interval. This has historical (and computational) reasons. We are currently working on additional nodes for the Time Series plugin and the moving average node will then allow you to choose if you want to compute the average left or right shifted or centered (which is what you want, I guess).

Michael

3

Thank you very much for this update…

I will stay tuned (esp. after the user training this week).

Best regards,

Jürgen

4

Sorry to resurrect this old thread.

I guess there were improvements in the moving average node since then, as it already has a lot of methods.

Do you plan to add methods that use the distance between time points for example (maybe in a different node)? I mean if the data points are not equidistant it might be misleading the current usage of Moving Average. What do you think? (Maybe it is already available, just I missed it.)

Thanks, gabor