Time Series_Optimize Holt-Winters Model_Wrong Connection

Big_Burger

Hi there,

I am a beginner with RapidMiner and have run into the following problem. I am using the Holt-Winters model to make predictions from a time series. I started with applying the model to my data. Then I added a forecast validation by creating this process:


Inside the Forecast Validation it looks like this:



So far so good. The process works, I get results and a validation of the results. 
At this point I wanted to add an optimizer, which automatically determines the best parameters for the Holt-Winters model. This is where my struggle starts. I adapted my process by adding the function "Optimize parameters":


Then I inserted the "Forecast Validation" into the "Optimize parameters" function:


Inside the "Forecast Validation" it looks like this: 


Now when I run the process, I get the following error message":


I already tried rearranging the functions in various ways but I didn't manage to solve the error. Please find attached the RapidMiner process. I really appreciate any help! 

Thanks in advance and best regards,
Simon

David_A

Hi @Big_Burger ,

please take a look at my attached process. I've adopted your solution with an example data set. The trick is to connect the time series model to the generic "out" port inside the Optimize Grid operator and also get it from the same output port in the main process "outside" the operator (you can see, that this output port then also is green, indicating a model is connected to it).

I hope this solves your problem.

Best regards,
David

David_A

Hi @Big_Burger ,

timeseries forecast models are of a different class than regular machine learning models.

Your process layout looks good, you only to connect the model output from the forecast validation to a normal "out" port (like you did with the test example set in your last screenshot). Afterwards everything should work fine.

Best,
David

Big_Burger

Hi @David_A,

Thank you for your feedback. I tried your solution, but it did not work (message "no result from optimization"). Nevertheless, my colleague solved the problem. I have to use a workaround (2 processes): the "apply forecast" function and the model connections have to be completely removed. In this way, the process yields the best parameters for my model. 
These parameters I can then use in my first process, to get the results. 

Maybe there would be a way to integrate both into one process, but I haven't figured it out yet.

Thanks anyway and best regards,
Simon

David_A

Hi,

I'll take a look at your process and check where the issue might be. But glad to hear that you found a working solution for now.

Best,
David

Big_Burger

Hi @David_A, 

Thank you for your answer. It works with your process! Sometimes the solution could be so easy. Thanks again for your help!

Best regards,
Simon