"How to speed up logistic regression?"

UsernameUsername Member Posts: 39 Maven
edited May 2019 in Help

the logistic regression operator has several parameters. I'm currently using it with the default values. I have ~150 attributes and ~300 000 examples. What parameters can speed up the learning process without decreasing the regression quality too much?



  • stereotaxonstereotaxon Member Posts: 10 Contributor II
    Not to disrespect RapidMiner, but if you might want to try R for the analysis. 

    make sure you have lots of memory, and fast memory at that.

  • TobiasMalbrechtTobiasMalbrecht Moderator, Employee, Member Posts: 294 RM Product Management

    since logistic regression uses an evoluationary approach to find an optimal solution you can reduce the runtime by specifying smaller values for the parameters [tt]max_generations[/tt] or [tt]population_size[/tt]. However you have to be aware, that it is more unlikely to find an optimal solution the more you put constraints on the parameters.

    Besides, there are to Weka operators that use other model fitting approaches. Maybe you might want to try these, so you do not have to leave RapidMiner completely! ;)

  • UsernameUsername Member Posts: 39 Maven
    The W-Logistic operator from WEKA is indeed much faster :).
  • IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder

    and I want to give a hint to our Kernel logistic regression approach "MyKLRLearner" which in my experience is again faster  ;)

Sign In or Register to comment.