I’m running an ARIMA prediction within a group loop that loops 242 times and outputs individual files. When test run individually, files process as expected; each file contains 396 rows and 14 columns. However, when the prediction model is placed within the loop, the ARIMA Lerner node fails execution after only 6 iterations.
I’m given this error message from the ARIMA Learner node: “Execute failed: illegal state: maximal count (2,147,483,647) exceeded: evaluations”
I’ve already modified the knime.ini file to have the Xmx size increased from 8048 to 16096, with zero effect, so I reverted back to 8048.
I would think that every time the loop started over again, the ARIMA Learner node would be reset and rerun and never get close to a processing limit.
Is there a way to avoid this or “clear the cache” for each loop?
This is an interesting one; can you upload your workflow in progress along with some sample data? At first glance I would guess the loop is somehow misconfigured and scaling wildly, but it’s also possible there’s a bug of course.
But it’s hard to know without seeing what you’re doing.
I talked to one of our data scientists about this one a bit more.
That error message reflects that the optimizer is failing to converge before the max 32bit signed number value kills it. (That’s an unclear error message, we should try and catch it and change it on our side.)
It might be interesting to see if you can isolate which group caused the error - and see if it errors outside the loop too.
It’s possible this might be fixed by changing parameter choices in the ARIMA. For example, if this is occurring due to a crazy exponential growth or something, adding a higher d value could address this.
Hi Scott. Thanks for the reply. My workflow executes as I want it to when I feed single files through it, but I have 337 groups to forecast so I put it in a loop.
I should have thought to put a screen capture in the original post, so here it is.
For the parameter choices, I’m attempting to add a Parameter Optimization Loop to the ARIMA model, since I had a similar thought, but I cannot get it to go. I get an error that states “Can’t merge FlowVariable Stacks! (likely a loop problem.)” I am getting this error even after creating a component from the ARIMA nodes (as suggested from another thread in this forum). I cannot figure out how to nest this loop inside the Group Loop.
