[python-package] Expose ObjectiveFunction class
#6586
Conversation
neNasko1
requested review from
guolinke,
jameslamb,
shiyu1994,
jmoralez,
borchero and
StrikerRUS
as code owners
|
@jameslamb, I'm reaching out to you as I noticed that you've tagged this as 'in progress'. I would appreciate it if you could share your thoughts on this change and its relevance to the project. If there is anything more to be done for this PR, I will be happy to contribute further. |
|
You will have to be patient with us... there is a small group of mostly-volunteer maintainers overseeing this project. This is quite a significant PR, as it greatly expands the public API of the Python package and the C API. It will be a while before I personally am able to provide a thoughtful review. Maybe others like @shiyu1994 or @guolinke will be able to help. |
| ], | ||
| ) | ||
| @pytest.mark.parametrize("num_boost_round", [5, 15]) | ||
| @pytest.mark.skipif(getenv("TASK", "") == "cuda", reason="Skip due to ObjectiveFunction not exposed for cuda devices.") |
There was a problem hiding this comment.
Why couldn't this also be exposed for the CUDA implementation?
There was a problem hiding this comment.
It segfaults on the CI tests, and I cannot build the CUDA version on MacOS.
There was a problem hiding this comment.
Where exactly does it segfault? 🤔 it seems like this should work 😅
Thank you for the response and sorry for the over-eagerness. I was just opening the door to discussion. |
|
No problem, we're happy to have you here contributing! In the interim, you might also consider picking up something smaller and more contained, like #6361. Once you have 1 commit merged to |
neNasko1
changed the title
ObjectiveFunction class to the Python APIObjectiveFunction class
| ], | ||
| ) | ||
| @pytest.mark.parametrize("num_boost_round", [5, 15]) | ||
| @pytest.mark.skipif(getenv("TASK", "") == "cuda", reason="Skip due to ObjectiveFunction not exposed for cuda devices.") |
There was a problem hiding this comment.
Where exactly does it segfault? 🤔 it seems like this should work 😅
|
@neNasko1 Please add
as it was suggested by @borchero . |
Provides a way to invoke the builtin functions outside of training. Implements access to gradient computations and output conversion methods.
This is a minimal implementation of #6585.