Viewing results of run operators when an error occurred at the end

mansour

I have run heavy-duty rapidminer models for a few days, at the end I see there was an error with writing the Excel file. Do I have to restart all the processes again? Are there any ways which I can see the results of the modelings or run the process from the point it faced with an error?
Regards.
Mansour

varunm1

Hello @mansour

That's a good question, I do face this issue sometimes and it's really hectic to run the process again. Let us see if there is any viable solution for this rather than process rerun. I am not sure if there is anyway other than storing results before writing into excel file. 

@mschmitz or @IngoRM any comments on this?

Thanks
Varun

sgenzer

yes I generally recommend storing parts of your process along the way for large runs. Standard practice.

mansour

Hi sgenzer
I am storing the results of model performance into an excel file when the model faces an error; before it, there is no result to save.

varunm1

Hello @mansour

The storing informed here refers to the use of store operator, to save the results in rapidminer repository. 

MartinLiebig

Hi Guys,

here is the bit longer answer:

How does an operator work?

Basically each operator has one method, to do it's thing (called doWork). Typically doWork methods are like this:

1. Get Data From Ports

2. Do Something With it

3. Deliver result to output ports.

So this means, that if we hit an error in (2), we cannot display much. The only way of doing this, would be to preserve the data when getting it from an input. That essentially means, that we need to do:

1. Get Data from Ports

1b. Store a security copy of the data

2. Do something with it

3. if it succeded, deliver result to output ports

3b. Else, use the security copy for something

Obviously this either takes a ton of more RAM or you need to store the backup to disc. That creates disc i/o and is thus also kind of expensive. Maybe our engineering team can do some magic by doing it in a seperate thread or so, but it will always cause some negative performance side effect.

That explains why this feature is not (yet) in. But we discuss solutions for this internally.

Best,

Martin

P.S Keep in mind, I'm a DS and not part of the RM Eng team. I might be mislead here.

Marco_Boeck

Hi,

I want to add one thing: If your process is running in the main Studio UI, you can also right-click a port of a finished operator and select to view the data from it. This even works if a later operator had an error. There is one caveat: If you're running low on memory, this temporary storage is purged. In that case, the data is not there anymore.

Regards,
Marco