"Modular Neural Networks ?"

marcelvanvelzenmarcelvanvelzen Member Posts: 1 Contributor I
edited May 2019 in Help

Is it possible to specify more modular Neural Networks in RapidMiner ? For example, the 4-bit parity problem is best solved by splitting the problem into two examplesets and each having their own learner. So, in total there are three neural networks working together.

If possible, how can I specify this ?




  • Options
    IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder
    Hi Marcel,

    sure. You could build two example sets by "copying" the data set with the IOMultiplier (don't worry, the data is actually not copied but only a view on it) and apply different ExampleFilter operator so that the data is splitted into the two desired example sets. Then you can apply two learners on both data sets on their own. You can now create predictions of the models and join them with the ExampleSetJoin operator. Then change the predictions to regular attributes with the ChangeAttributeRole operator and learn the final model on it.

    It's a bit tricky but it should be possible. Of course, you could also write your own operator performing all these steps at once but I actually would always prefer this more modular way of working since you can more insight and control over the subprocesses.

Sign In or Register to comment.