chi-squared statistics returned by RapidMiner

Legacy UserLegacy User Member Posts: 0 Newbie
Hello,

I tried to use Rapid Miner to compute Chi-squared weights of a set of attributes.

Chi-squared weights are locally defined, i.e. computed with respect to a given class, and the local weights of an attribute are sometimes combined to give a single global weight of the attribute (e.g. by summing up all of its local weights, and taking its max local weight).

My question is, how is the single chi-squared weight returned by RapidMiner computed?

Thank you very much.

Best Regards
CW

Answers

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

    all class based weights are summed up.

    Cheers,
    Ingo



  • ironfrownironfrown Member Posts: 2 Contributor I

    I have been chasing this up for while as well. I am still not clear on your answer.

     

    How do you derive Weight from Chi-Squared for two variables X and Y? I imagine we construct a contingency table of "r" rows (corresponding to all classes of variable X) and "c" columns (corresponding to all classes of variable Y). I have no issue calculating Chi-Squared statistic of X and Y independence. How do you get the weight for X and Y from this?

     

    Jacob

  • Fred12Fred12 Member Posts: 344 Unicorn

    I would be interested in this, too...

    any way to look into the source code?

  • mmichelmmichel Employee, Member Posts: 129 RM Engineering

    The core of RapidMiner Studio is open source and available at github: https://github.com/rapidminer/rapidminer-studio

     

    If you want to get a better understanding of the 'Weight by Chi Squared Statistic' operator, you can find the implementation details here.

  • User42010User42010 Member, University Professor Posts: 5 University Professor
    I would like to know the formula used to calculate the weights generated by chi-square.

    Hope to get help concerning this. 
Sign In or Register to comment.