Options

Running multiple regression methods on multiple problems

rapiduser77rapiduser77 Member Posts: 2 Contributor I
edited November 2018 in Help
Hi,

I would like to run multiple regression methods on multiple problems. Is there anyway to choose all the regression methods that I would like to use and problems and just run rapidminer once returning the results?

It is not hard to do this for one method on one problem, but if you have many, it requires time to run one by one.

Answers

  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    Hi,

    unfortunately there is no easy automated way of trying all different algorithms with different parameters etc., so you have to invest the manual work.

    You could try a combination of Select Subprocess, Loop Parameters (to try all the different subprocesses), and the Log operator, but to my experience the effort of creating and maintaining the process is not worth it.

    Best regards,
    Marius
  • Options
    rapiduser77rapiduser77 Member Posts: 2 Contributor I
    If this is not possible, what about doing this in code, java?

    I saw a couple of webpages explaining the usage of rapidminer as a java library.  There are some examples, but I couldn't find a proper resource about how to perform a regression task.

    Is there any document explaining that?
  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    On rapidminer.com under Support -> Documentation there is a whitepaper called "How to extend RapidMiner". This does not explain how to use RapidMiner as a library, but it helps you to get the grips on the RapidMiner API. To use the software as a library you can use the rapidminer.jar that ships with the installation, or add the project to eclipse, which should be explained in the whitepaper.

    Please remember that RapidMiner is released under the AGPL, which means that your software that links to RapidMiner must also be licensed under a compatible license.


    But before going the step of writing Java code lets consider some other options. I would probably prepare a set of processes (one per algorithm), and create one master process that executes all the other processes with the Execute Process operator.
    The master process would load the data once and pass it to the subprocesses via the process inputs.

    Best regards,
    Marius
Sign In or Register to comment.