probabilistic relational data mining

lindawu

Have anyone have experience with probabilistic relational data mining? I want to build a model to learn rules from the data, for example:

if A&B&C => D, with probability > 90%

Is there any method in RapidMiner can do this?

land

Hi,
nothing beyond the rule learner. But if you want to add such a feature, you are welcome

Greetings,
Sebastian

lindawu

Is it possible I use the current rule learner in RapidMiner, but add a new criteria, for example, instead of using information gain, I can use probability > 90%? If I want to write my own criteria, how can I do that?

land

Hi,
you simply write a class extending com.rapidminer.operator.learner.rules.AbstractCriterion. This must then be added to the arrays of Classes and Names in the RuleLearner operator's class file.

Unfortunately this is not extend-able in the current version, so you so you cannot simply add a new one. But we might add a mechanism for Extensions to register own criteria. I have this already on my list. Then you could simply build an extension, that registers an additional criterion. Currently you will have to exchange one of the old criteria with your one by modifying the array entries.
For getting an tutorial how to write a RapidMiner Extension, please see our web shop.

Greetings,
  Sebastian