Skip to Main Content

Data Science & Machine Learning

Announcement

For appeals, questions and feedback about Oracle Forums, please email oracle-forums-moderators_us@oracle.com. Technical questions should be asked in the appropriate category. Thank you!

Interested in getting your voice heard by members of the Developer Marketing team at Oracle? Check out this post for AppDev or this post for AI focus group information.

Oracle 21c XGBoost "feature not supported"

user8005280Oct 15 2021

XGBoost is a popular and award winning data mining algorithm.
Oracle has implemented it on Oracle 21c by way of an external module. (I think there is an R module behind the scenes). Oracle has re-used and integrated the open source code.
As such, XGBoost is perhaps somewhat different under-the-hood, than other Oracle data mining algorithms.
I have been running data mining algorithms in Oracle 19c (primarily using SVM and random forests) and have just upgraded to 21c and wanted to give XGBoost a try.
I am getting a strange error when running the CREATE_MODEL procedure.
ERROR at line 1:
ORA-40216: feature not supported
ORA-06512: at "SYS.DBMS_DATA_MINING", line 355
ORA-06512: at "SYS.DBMS_DATA_MINING", line 605
ORA-06512: at "SYS.DBMS_DATA_MINING", line 564
"Feature not supported" is quite vague and there is not much information on it.
It could be anything ranging from formatting and exporting the data to the R procedure, to some internal bug in the R procedure, to a whole host of other things.
I do know that the error occurs right away, within one second of calling CREATE_MODEL.
I am using quite a number of DM_NESTED_BINARY_DOUBLES objects in my data table. I don't know if that could be the issue.
If I had access to the Oracle source code, then I could probably find the error myself just by inspecting the code. Some programmer at Oracle has an "if statement" there somewhere and raises that error when a certain condition is not met. The problem would be so easy to solve if I had access to that source code. Probably take me 20 minutes to figure it out. But I do not. It is just this big black box.
Anyway, rather than porting my existing code, that works fine for Random Forests, I am going to have to create an extremely simple stand alone test case -- probably use a data table with 10 columns and 100 rows, and see if there error persists. Probably not. Then incrementally build up the complexity until I can reverse engineer the error. This is not the most efficient way to do things. What could have been 20 minutes is instead 3 or 4 hours. And that is only to find out what the bug is. Then to fix it, who knows.
Sorry for the rant. A generic error such as "feature not supported" simply gives me no information. Whoever programmed that should have done a better job with that error message.
Thank you,

Comments

Post Details

Added on Oct 15 2021
3 comments
288 views