[jvm-packages] change DeviceQuantileDmatrix into QuantileDMatrix

[wbo4958]

To fix #8453

[trivialfis]

Please mark the change as breaking.

The new QuantileDMatrix accepts both CPU and GPU based inputs, can we implement it on JVM side as well?

[wbo4958]

So you mean we also change the CPU way to build QuantileDMatrix instead of plain DMatrix?

[trivialfis]

Yes, similar to pyspark interface, we can have QuantileDMatrix as default for both hist and gpu_hist.

[wbo4958]

Ok, I will file another PR for it.

[wbo4958]

@trivialfis please help to check the CI why the xgboost-gpu tests run on windows and mac?

[trivialfis]

There's an undefined symbol: XGQuantileDMatrixCreateFromCallbackImpl when compiling without CUDA.

[wbo4958]

Thx, fixed it.

But I'm still curious why the test can pass in ubuntu? and I locally build CPU version with USE_CUDA=OFF, and it still pass

[trivialfis]

Thank you for the quick fix! triggered the CI.

[trivialfis]

@hcho3 Could you please help check whether the CI is working alright?

[hcho3]

@trivialfis An EC2 worker running Windows got stuck and hung for 10 hours. I had to manually kill it.

[wbo4958]

@trivialfis please help to trigger the CI. Thx

[trivialfis]

Done.

[wbo4958]

@trivialfis Seems the CI is not triggered.

[hcho3]

I triggered it again.

[wbo4958]

Does the python test have some issues?

E           ImportError: /opt/python/envs/cpu_test/lib/python3.8/site-packages/_cffi_backend.cpython-38-x86_64-linux-gnu.so: symbol ffi_type_uint32 version LIBFFI_BASE_7.0 not defined in file libffi.so.7 with link time reference

[trivialfis]

Could you please remind me what's the reason behind setters not being supported?

[wbo4958]

this file is renamed by DeviceQuantileDMatrix which is not supporting setter before (when the PR was merged). Will file another PR to support setter if it indeed has supported that.

[wbo4958]

@trivialfis @hcho3 please help to trigger the CI. Thx

[hcho3]

Done

[wbo4958]

@hcho3 @trivialfis please help to trigger the CI again, I just merged the latest commits into this PR. Thx

[wbo4958]

@trivialfis please help to check why python test failed in this CI, I suppose this PR won't touch any python code.

[hcho3]

@wbo4958 I re-created the machine env and re-triggered the tests. The CI is passing on the master branch, so hopefully it will remove the error.

[wbo4958]

Thx @hcho3, @trivialfis please help to review again. Thx