How to ARIMA in 5.0

[PChott]

Has anyone create any ARIMA model in RapidMiner 5.0?

I'm searching web but did't find any good result 

[Sebastian Land]

Hi,
actually we do things a little bit different (and more flexible) in RapidMiner: Instead of building a certain algorithm for processing time series as ARIMA, we transform time series to normal data sets by windowing them. After this is done, a standard learning algorithm (linear regression, SVM, neural net) can be applied to the problem. Of course you can also apply moving averages. Together this one solves then the ARIMA problem, but with the advantage that you can optimize each single step and replace LinearRegression as in original ARIMA by SVM...

Greetings,
  Sebastian

[PChott]

Thx for information.

Due to I need ARIMA just for comparision towards other methods in RapidMiner I'll try to push data to R script at the moment.

Regards,
Jan

[Marin]

To refresh it,

did anyone create an ARIMA operator for RM?
Or a process emulating it?
Or a process using R script so ARIMA can be ran concurently with other processes from within RM? (I have to compare ARIMA with different algorithms and I would like to do it in RM)

PS: ARMA, ARIMAX or similar will also be greatly appreciated.

Cheerz,

Marin

[wessel]

Hey,

What I do is simply write out the predictions to a file and process them later.
Gives you a clear indication of what is going on and allows you to use a fast ARIMA implementation.
Managed to use the R script operator to call ARIMA, but this can be really slow.

Best regards,

Wessel

[Marin]

Indeed, in my case integration of R is also running slow. For the HoltWinters to just take a column from RapidMiner, create model, calculate predictions and return results it takes 10 minutes, while native R needs about 2 seconds (it imports from a file).