exporting processes with Weka operators in the RM 58

Question

I installed the latest version of RM (5.0.08). For some reason I could not export processes using Weka operators. I reinstalled RM, the same problem appeared. Finally I reinstalled a previous version (5.0.05), and exporting worked properly.

Regards,
Dan

IngoRM · Answer

Hi,
For some reason I was not successful the first time. It would still be nice for your business customers to have this operator - in particular for those which use imperfect/missing data frequently.

I completely agree. By the way: You delivered every single reason why people should really consider to change to the Enterprise Edition of RapidMiner  ;D* You have a problem with a process: Don't worry, ask our Enterprise Support and we will help you in resolving your problems immediately.
* You need another operator or an extension of existing ones: Don't worry, we are happy to fulfill our customer needs as soon as possible.

I usually try not to bother you guys with too much advertisement for our Enterprise Services here in the community forum but - hey - this was a great opportunity  ;)

Cheers,
Ingo

dan_agape · Answer

Hi Ingo,

Thanks - I'll try to use that solution again. For some reason I was not successful the first time. It would still be nice for your business customers to have this operator - in particular for those which use imperfect/missing data frequently. Just for productivity and user friendly-ness reasons.

Many thanks for making this amazing software available! It really makes a difference for the Data Mining community.

Best,
Dan

IngoRM · Answer

Hi,
Obviously RM is already quite strong in preprocessing functionality, but some essential operators, which already exist in Weka, have been forgotten.

it's not a matter of "forgetting". In our own projects, we did just had not the need for these operator or - more likely - we had the ability to build those operators by using the other, already existing, operators. And since noone else paid us for implementing them...  ;)

In contrast to the learning schemes, which can easily be integrated from Weka, this is hardly true for the filters. Data structures are different for both applications and filtering changes data a lot (in contrast to mere modeling) and hence we would have to integrate and optimize each Weka filter anew - which needs the same amount of time as simply implementing them for RapidMiner and make use of the more powerful data structures and allow things like preprocessing models, views with on-the-fly calculation etc.
Someone suggested a non-straightforward solution which actually I could not use.

What's wrong with this solution? Works well for me...

Cheers,
Ingo