XGBoost and LightGBM implementations

Nirr3Nirr3 Member Posts: 8 Contributor II
Hello,

while the h20 gbm definitely is a performing model, I often find myself using the XGBoost or lightgbm model from either R or python instead. Given this is one of the most commonly used gbm models, any chance we could have them added to RapidMiner.

Tagged:
Andy3

Comments

  • sgenzersgenzer 12Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager
    Hi @Nirr3 - I'm going to connect you with @bhupendra_patil. He has a pilot project that is working on exactly this kind of idea.

    Scott

  • christos_karraschristos_karras Member Posts: 50 Guru
    Hi @sgenzer and @bhupendra_patil,

    Is there any news about this? I would also be interested in trying a RapidMiner XGBoost implementation and see how it compares with an existing model based on H2O Gradient Boosted Trees.

    Thanks
  • hughesfleming68hughesfleming68 Member Posts: 323 Unicorn
    Try the Smile extension. I have had very good results with it, especially the GBT.
  • jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    edited August 2020
    Xgboost is available for Java, including its GPU support. 
  • jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    I have also noticed that the current version of H2O provides not only GBM (I assume this model is available in RM) but also XGBoost interface (yes, the original one as well as its light version), with GPU support which gives fantastic training acceleration but which restricts the tree learning options, so I found out it is not producing models of the accuracy as good as that of the CPU based backend.
  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,280 RM Data Scientist
    Seems fairly new to me?

    Best,
    Martin

    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,280 RM Data Scientist
    If yes, then this is the problem: "Windows not supported in the JVM package"
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    @mschmitz Yes, this was the one, also announced on the main web site. Pity, looks like a short wait for Windows to be supported?
  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,280 RM Data Scientist
    Its like this since 2016 if i remember correctly. They have no interest in supporting Windows.
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    I've done some more digging and I found out this French fork of XGBoost. If anybody wanted to do a DIY xgboost extension for RM (which seems to include Windows), here is a link to try:
    It is becoming a sort of second-hand development and I am not sure if GPU would be supported in this project.
  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,280 RM Data Scientist
    i am aware of this. But adding self-build jars with platform dependency is a bit trickier than using normal things. I doubt that I can do this as a short side project without help of people like @jczogalla or @Marco_Boeck . And these folks are usually busy developing other cool new features. So we would need to push this through the normal product management cycles.

    Cheers,
    Martin

    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
Sign In or Register to comment.