(SOLVED) Wrong Correlation Matrix Output

ndf-lsndf-ls Member Posts: 8 Contributor II
edited September 2019 in Help
Hello,

i just started working with rapidminer and have a problem using a correlation matrix:
My database is an excel sheet. The feed into Rapidminer works, however when i connect a correlation matrix operator next to it, the results in the matrix show many "?" instead of values between -1 and 1. Single cells show actual values, however some of them are outside of the normal -1 to What does this mean? I have 15 attributes, is that to much?!

Thank you very much in advance
Tagged:

Answers

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

    15 attributes are definitely not too much. There is either a problem with the process setup, with the data or maybe there is a bug in the Correlation Matrix operator. Can you please post the XML of your process?
    It would also be good if you could provide a sample of your data.

    Best regards,
    Marius
  • ndf-lsndf-ls Member Posts: 8 Contributor II
    The problem is now solved. The reason for the wrong Output is the following:
    The used Database consisted of many constant values for most of the attributes. The "Correlation Matrix"-Operator uses the Pearson Correlation Coefficient. As the standard deviation of this formula is listed in the denominator, the correlation is not defined for constant values. This was the reason for the "?" Output in many of the correlation matrix cells.
    Big thanks to Marius for his great support!
  • MruduMrudu Member Posts: 1 Newbie
    Hi, I am facing similar problem with 15 attributes. You explained the reason, can you please tell the solution also. I am attaching mt process xml file for your reference.
  • sgenzersgenzer Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager
    hi @Mrudu this is a thread from 2014. I'd highly recommend starting a new discussion :wink:

    Scott

Sign In or Register to comment.