Process Failed - Holt-Winters is not ... trained before

kose

Hi all, 

I am pretty new at using rapidminer and I have a question which may have a very simple solution. When I run a pre-made process, I am getting this error:

"Process failed. Holt-Winters is not trained on a time series Run initializeSmoothing(List<Double) values before." 
prnt.sc/1l476bd

 The process is working on the computer who sent me this process even though we follow every step exactly the same. I have installed Forecasting, Operator Toolbox, Holt-Winters Filtering extensions. 

Thanks in advance!

tftemme

Hi @kose,

You probably found a bug, which rarely occurs and is caused by specific structure of your input data on which you want to train the Holt-Winters model on.
We will work on the bug and fix it in a future version of RM Studio.

For now, you can check if you input data set contains one or more small period of 0 values. If the number of consecutive zeros in your data set is larger than the period, the bug is triggered. In this case please try a larger period for the Holt-Winters.
Otherwise it also can be that you have every x value a zero, you have selected multiplicative for the seasonality model and you period is x (for example you have daily data but always 0 for the sunday (so every 7.th value) and you have 7 as the period. Than the error can be triggered as well. Than you can change to additive seasonality model for now.

Hopes this helps,
Best regards,
Fabian

kose

I am still having this problem. I also tried on another computer and I got the same error. I'll be thankful for any help!

MartinLiebig

Hi,

this is something I never saw before. Any chance you can share the data and the process so that we can have a look?

Cheers,

Martin

tftemme

Hi @kose,

You probably found a bug, which rarely occurs and is caused by specific structure of your input data on which you want to train the Holt-Winters model on.
We will work on the bug and fix it in a future version of RM Studio.

For now, you can check if you input data set contains one or more small period of 0 values. If the number of consecutive zeros in your data set is larger than the period, the bug is triggered. In this case please try a larger period for the Holt-Winters.
Otherwise it also can be that you have every x value a zero, you have selected multiplicative for the seasonality model and you period is x (for example you have daily data but always 0 for the sunday (so every 7.th value) and you have 7 as the period. Than the error can be triggered as well. Than you can change to additive seasonality model for now.

Hopes this helps,
Best regards,
Fabian

kose

Hi @mschmitz and @tftemme,

I will try it again with these instructions. If I find a solution, I will let you know as soon as possible. Otherwise, I can share the data and the process. 

Thank you so much for your help. 

Regards,
Kose

MartinLiebig

Hi @kose ,

thanks for the offer. I think we were able to reproduce the error on test data.

Please keep us posted on your progress.

Cheers,

Martin

kose

Hi @mschmitz @tftemme,

Yes, it works! As you said, the problem was about the data, if the input data has a period full of 0's, then it gives this error. I appreciate your help so much.

Regards,
Kose