The past couple of weeks have been a tsunami of EPM for me. I first attended Ascend26 in Orlando where I gave two presentations, one on Narrative Reporting and another on Advanced Predictions. Both went well and I was feeling pretty good. Then after a brief two nights at home, I jumped on a plane for Denver to attend Kscope26. I did the same two presentations there, and again I felt that both went well. However, due to the honest feedback of an attendee, I learned that I had some incorrect information in my Advanced Predictions presentation.
Clarification #1
I commented in the presentation that the available algorithms included a recently added one called ETF. I believe it showed up in the 26.05 (May) release, but possibly in 26.04 (April). However, it is no longer in the selection list of algorithms. I assume there were issues with it and Oracle removed it until a resolution is implemented.
Clarification #2
This is the bigger mistake on my part. When I talked about the available algorithms, I stated that AutoMLx selects the best algorithm based on your dataset. While true that it does select the best, the clarification is that it does not select Light GBM, XGBoost, Prophet, or SARIMAX. These multivariate algorithms must be specifically selected to be used in the prediction. AutoMLx selects from several other algorithms, and importantly, these other ones are univariate algorithms. So, if you want to use AutoMLx, be aware that it is not using multivariate algorithms. I believe it is still a valuable option to use AutoMLx, but it is a bit misleading. Oracle documentation does not clearly highlight this, but that really is no excuse on my part, I should have done deeper analysis and caught it sooner.
My apologies for the errors in my presentation on Advanced Predictions. While this does not diminish the usefulness of Advanced Predictions or damper my enthusiasm for it, it does serve as a great reminder to not fully rely on the documentation, take the time to validate what the documentation is stating. Reach out to others, including Oracle, and ask questions or get clarifications. Use Cloud Customer Connect to see what other users are experiencing and how they are using/testing the models. One of the great things about the EPM community is the willingness of so many to share their knowledge and experiences, all you have to do is ask.
As always, happy EPM’ng!
P.S. Special shout out to Patrick Northcraft for sharing his knowledge with me, it is greatly appreciated.
Random EPM Notes 