"Export PMML or SQL"

Dan

Hi there,

I am new to RMiner. I wander if it has any facility of exporting models in XML or SQL formats. These facilities are very useful allowing exchange of models generated with a DM package with another one or using them within a database, or simply sending models to users.
Any idea about these? If they are not yet available - can the providers give an indication if these are to be included in the near future?
Thanks.

Dan

land

Hi Dan,
our models are already stored in XML, as does everything else. I admit, its a rather generic XML format, but nevertheless its XML Just use the IOObjectsWriter and deacitved the zip compression.

Greetings,
  Sebastian

Dan

Thanks Sebastian. I have tried to use a model built by RapidMiner in XML format in Clementine just to test the portability of the models. I produced a decision tree, exported it in a file in XML format (and not the default zipped XML) using IOModelWriter but also IOObjectsWriter. Unfortunately I got the message "the file has the wrong type" from Clementine in both cases. Is portability of models documented somewhere with RapidMiner? This would be an interesting feature as Clementine models can be employed without using the expensive Clementine client or server, but in a lighter environment, if saved in XML (that is, using PMML).

Best
Dan

land

Hi Dan,
unfortunatly our XML format is not PMML, just a XML format. It is more generic and covers all available model types and learning algorithms. This is the major draw-back of PMML, because it is limited on only a handfull of models. Since both formats are XML, one perhabs could use XSLT to transform one format into the other, but I don't know PMMLs specifications.
PMML will be probably supported after it entered its finite state. But since versions change rather quickly and the big companies of the standard group enforce this quick developement, this might last some time.

But I feel a little bit proud, that you want RapidMiner to learn the models and Clementine to do the dump application. Or did I get you wrong here?

Greetings,
  Sebastian

Dan

Sebastian, thanks for clarification. Indeed, it may be an alternative to move to RM for obvious reasons (reduced costs) and still be able to reach DM consumers usually from Clementine's and other products' markets (SPSS propose light environments to deploy the models in PMML, to be used without possessing Clementine). If Clementine v12 got the actual PMML then probably some continuity is expected, so perhaps this may be considered by RM at some earlier moment than currently estimated - hopefully.

Anyway, can you please give some examples of packages where the models produced with RM and saved in XML can be employed/deployed currently? Do you plan to add some operator to convert a model representation to SQL?


Regards
Dan

TobiasMalbrecht

Hi Dan,

to clarify things further: as far as we are aware of, the XML model descriptions RapidMiner (RM) produces when saving models are only read by RM. The main intention of the format was to persistantly store models. Principally, the XML could be read by other applications but that was not the primary focus. Of course, a generic language for exchanging data mining models such as the PMML is a noble and useful thing and we surely keep observing the developments in that area. Unfortunately, in the short to medium term we will however not be able to support PMML in RM since there are a lot of other things on our schedule which will surely improve RM. But in the long run, supporting PMML might definitely be an option.

One thing we have in mind is a better support of interconnecting RapidMiner with databases. A complete model formulation in SQL is unfortunately not possible yet. But we definitely plan to work in that area for the next major release to allow a better and closer integration of RM in the database world. Apart from that, RM can already load and store data from databases and execute arbitrary SQL statements. Hence, your deployment tasks might already be executable by RM completely. If you would like to have assistance in deploying RM or integrating RM into your infrastructure, just send us a mail and we can talk about in how far we are able to help you. Maybe even customising RM for your needs might be an option, as this might be even cheaper than obtaining a standard SPSS license!

Regards,
Tobias

Dan

Tobias, thanks. It's a clear picture now. I am interested and started to study RM's capabilities. Thanks for the offer, at the appropriate moment we can contact your company.

Best,
Dan

TobiasMalbrecht

Hi Dan,

sure, just drop us a note. If you have also more questions concerning the capabilites of RM, feel free to ask ...

Regards,
Tobias

mz

Dan,

Zementis(www.zementis.com) is part of the PMML standard and we have contributed to the PMML export for R, another open source package.  We strongly believe in open standards and support the easy deployment of PMML models with our ADAPA predictive analytics decision engine.  In case you are looking for a cost-effective, highly scalable deployment environment, ADAPA is available on the Amazon cloud computing infrastructure as a SaaS option (no fixed cost, private & secure instances).

We are currently considering to contribute similar PMML export functionality to RapidMiner.  It would be very helpful for us to know what potential users are looking for, so we can prioritize and focus on adding the right export capabilities.  What type of algorithms would you be interested in for the PMML export?  From your posting, it looks like Decision Trees would be first on the list.

Your feedback is highly appreciated!

Kind Regards,
Mike