"Lower accuracy of libsvm in rapidminer vs weka with same parameters"

njasajnjasaj Member Posts: 18 Contributor II
edited June 7 in Help
Hi,
I am really getting confused. I used a same data set with same parameters in weka and rapidminer for classification with libsvm. the strange result was that weka accuracy for prediction was 15 percent higher than rapidminer???(I used same C and same kernel and same gamma for rbf kernal).I think there is something wrong with weka. it's accuracy on training set for testing phase is 100%.
Can any body explain it?
Thanks a lot.
Tagged:

Answers

  • njasajnjasaj Member Posts: 18 Contributor II
    In addition to above i must say that my data set is imbalanced. rapidminer fail to predict the two smaller classes and just predict the largest class
    725  188    211
    3      1          1
    1      0            1

    but weka with same parameters gives
    729    0    0
    124    65    0
    111    0    101
    I did not use any weight for classes in weka.Help me please and tell me what is wrong with rapidminer?

    thank you.
  • MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869   Unicorn
    There are many different implementations of the SVM. If the Weka implementation works better on your data, you can easily integrate Weka with RapidMiner by installing the Weka Extension. For comparing models please remember to cross-validate them - as you know the training error alone is not an indicator on how well the model will work on unseen data.

    Best, Marius
  • MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869   Unicorn
    BTW, you should optimize each implementation separately - for different implementations different parameter sets *might* work better.
  • Fred12Fred12 Member Posts: 344   Unicorn

    does  the Weka LibSVM implementation still exist or was it removed? Somehow I cannot see any LibSVM Weka implementation anymore in the list...? Would like to try it out..

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,761   Unicorn

    The version of Weka we implement is v3.6.9. But from this: https://weka.wikispaces.com/LibSVM it appears it's a 3rd party tool that needs to be downloaded and installed seperately. It looks like if you want to use the Weka implementaiton of LibSVM, you have to download it extra. 

  • Fred12Fred12 Member Posts: 344   Unicorn

    thanks, but it seems that the download links are broken..

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,761   Unicorn

    I can't do anything about that. :( Maybe contact them, somehow?

Sign In or Register to comment.