1822 Commits

Author SHA1 Message Date
Jiaming Yuan
4b10200456
[coll] Improve event loop. (#10199)
- Add a test for blocking calls.
- Do not require the queue to be empty after waking up; this frees up the thread to answer blocking calls.
- Handle EOF in read.
- Improve the error message in the result. Allow concatenation of multiple results.
2024-04-18 03:29:52 +08:00
Jiaming Yuan
1022909bbe
Fix global config for external memory. (#10173)
Pass the thread-local configuration between threads.
2024-04-11 01:29:28 +08:00
Jiaming Yuan
f0a138f33a
Fix pyspark with verbosity=3. (#10172) 2024-04-09 23:18:56 +08:00
Jiaming Yuan
8bad677c2f
Update collective implementation. (#10152)
* Update collective implementation.

- Cleanup resource during `Finalize` to avoid handling threads in destructor.
- Calculate the size for allgather automatically.
- Use simple allgather for small (smaller than the number of worker) allreduce.
2024-03-30 18:57:31 +08:00
Jiaming Yuan
230010d9a0
Cleanup set info. (#10139)
- Use the array interface internally.
- Deprecate `XGDMatrixSetDenseInfo`.
- Deprecate `XGDMatrixSetUIntInfo`.
- Move the handling of `DataType` into the deprecated C function.

---------

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
2024-03-26 23:26:24 +08:00
Hui Liu
ff549ae933 sync upstream code 2024-03-20 16:14:38 -07:00
Jiaming Yuan
ca4801f81d
Work with IPv6 in the new tracker. (#10125) 2024-03-20 05:19:23 +08:00
Jiaming Yuan
53fc17578f
Use std::uint64_t for row index. (#10120)
- Use std::uint64_t instead of size_t to avoid implementation-defined type.
- Rename to bst_idx_t, to account for other types of indexing.
- Small cleanup to the base header.
2024-03-15 18:43:49 +08:00
Jiaming Yuan
56b1868278
Fix compilation with the latest ctk. (#10123) 2024-03-15 08:04:41 +08:00
Hui Liu
968dbf25fb merge latest changes 2024-03-12 09:13:09 -07:00
Jiaming Yuan
1450aebb74
Fix pairwise objective with NDCG metric along with custom gain. (#10100)
* Fix pairwise objective with NDCG metric.

- Allow setting `ndcg_exp_gain` for `rank:pairwise`.

This is useful when using pairwise for objective but ndcg for metric.
2024-03-11 14:54:10 +08:00
Jiaming Yuan
2c13f90384
Support graphviz plot for multi-target tree. (#10093) 2024-03-09 05:35:25 +08:00
Jiaming Yuan
e14c3b9325
Optional normalization for learning to rank. (#10094) 2024-03-08 12:41:21 +08:00
Jiaming Yuan
3941b31ade
Disable column sample by node for the exact tree method. (#10083) 2024-03-01 14:16:10 +08:00
Jiaming Yuan
8189126d51
Add CUDA iterator to tensor view. (#10074) 2024-03-01 14:15:31 +08:00
Jiaming Yuan
5ac233280e
Require context in aggregators. (#10075) 2024-02-28 03:12:42 +08:00
Jiaming Yuan
0ce4372bd4
Use UBJSON for serializing splits for vertical data split. (#10059) 2024-02-25 00:18:23 +08:00
Jiaming Yuan
2e4ea5ecc0
Support f64 for ubjson. (#10055) 2024-02-21 02:18:42 +08:00
Jiaming Yuan
d37b83e8d9
Fix UBJSON with boolean value. (#10054) 2024-02-20 22:13:51 +08:00
david-cortes
6e3c899ba7
[R] Don't cap global number of threads for serialization (#10028) 2024-02-20 11:13:00 +08:00
Louis Desreumaux
edf501d227
Implement contribution prediction with QuantileDMatrix (#10043)
---------

Co-authored-by: Jiaming Yuan <jm.yuan@outlook.com>
2024-02-19 21:03:29 +08:00
david-cortes
a730c7e67e
[R] allow using seed with regular RNG (#10029) 2024-02-04 16:22:22 +08:00
Hui Liu
44db1cef54 Merge branch 'master' into sync-2024Jan24 2024-02-01 14:41:48 -08:00
Philip Hyunsu Cho
4dfbe2a893
[CI] Test building for 32-bit arch (#10021)
* [CI] Test building for 32-bit arch

* Update CMakeLists.txt

* Fix yaml

* Use Debian container

* Remove -Werror for 32-bit

* Revert "Remove -Werror for 32-bit"

This reverts commit c652bc6a037361bcceaf56fb01863210b462793d.

* Don't error for overloaded-virtual warning

* Ignore some warnings from dmlc-core

* Fix compiler warnings

* Fix formatting

* Apply suggestions from code review

Co-authored-by: Jiaming Yuan <jm.yuan@outlook.com>

* Add more cast

---------

Co-authored-by: Jiaming Yuan <jm.yuan@outlook.com>
2024-01-31 13:20:51 -08:00
Hui Liu
2cb579ff3c fix memory type 2024-01-26 15:46:42 -08:00
Jiaming Yuan
a76d6c6131
Fix cpp deprecation. (#10010) 2024-01-26 02:13:40 +08:00
Philip Hyunsu Cho
c8f5d190c6
[CI] Stop Windows pipeline upon a failing pytest (#10003) 2024-01-24 22:54:21 -08:00
Hui Liu
3fe874078c merge latest changes 2024-01-24 13:30:08 -08:00
Hui Liu
069cf1d019 use __HIPCC__ for device code 2024-01-24 11:30:01 -08:00
Hui Liu
1e0ccf7b87 fix random 2024-01-21 12:48:41 -08:00
Jiaming Yuan
cacb4b1fdd
Fix gain calculation in multi-target tree. (#9978) 2024-01-17 13:18:44 +08:00
Jiaming Yuan
85d09245f6
Fix error handling in the event loop. (#9990) 2024-01-17 05:35:35 +08:00
Hui Liu
9759e28e6a compiler errors fix 2024-01-12 12:09:01 -08:00
Hui Liu
1e1e8be3a5 merge latest, Jan 12 2024 2024-01-12 09:57:11 -08:00
Hui Liu
c42c7d99f1 fix memoryType 2024-01-11 14:10:30 -08:00
Jiaming Yuan
2f57bbde3c
Additional tests for attributes and model booosted rounds. (#9962) 2024-01-09 09:54:39 +08:00
Jiaming Yuan
38dd91f491
Save model in ubj as the default. (#9947) 2024-01-05 17:53:36 +08:00
Jiaming Yuan
c03a4d5088
Check support status for categorical features. (#9946) 2024-01-04 16:51:33 +08:00
Jiaming Yuan
621348abb3
Fix multi-output with alternating strategies. (#9933)
---------

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
2024-01-04 16:41:13 +08:00
david-cortes
3c004a4145
[R] Add missing DMatrix functions (#9929)
* `XGDMatrixGetQuantileCut`
* `XGDMatrixNumNonMissing`
* `XGDMatrixGetDataAsCSR`

---------

Co-authored-by: Jiaming Yuan <jm.yuan@outlook.com>
2024-01-03 17:29:21 +08:00
Jiaming Yuan
a7226c0222
Fix feature names with special characters. (#9923) 2023-12-28 22:45:13 +08:00
Hui Liu
2d7ffbdf3d merge latest changes 2023-12-13 21:06:28 -08:00
Dmitry Razdoburdin
43897b8296
Sycl implementation for objective functions (#9846)
---------

Co-authored-by: Dmitry Razdoburdin <>
2023-12-12 14:41:50 +08:00
Jiaming Yuan
faf0f2df10
Support dataframe data format in native XGBoost. (#9828)
- Implement a columnar adapter.
- Refactor Python pandas handling code to avoid converting into a single numpy array.
- Add support in R for transforming columns.
- Support R data.frame and factor type.
2023-12-12 09:56:31 +08:00
Jiaming Yuan
42de9206fc
Support multi-target, fit intercept for hinge. (#9850) 2023-12-08 05:50:41 +08:00
Jiaming Yuan
39c637ee19
Use array interface in Python prediction return. (#9855) 2023-12-08 03:42:14 +08:00
Dmitry Razdoburdin
381f1d3dc9
Add support inference on SYCL devices (#9800)
---------

Co-authored-by: Dmitry Razdoburdin <>
Co-authored-by: Nikolay Petrov <nikolay.a.petrov@intel.com>
Co-authored-by: Alexandra <alexandra.epanchinzeva@intel.com>
2023-12-04 16:15:57 +08:00
Jiaming Yuan
8fe1a2213c
Cleanup code for distributed training. (#9805)
* Cleanup code for distributed training.

- Merge `GetNcclResult` into nccl stub.
- Split up utilities from the main dask module.
- Let Channel return `Result` to accommodate nccl channel.
- Remove old `use_label_encoder` parameter.
2023-11-25 09:10:56 +08:00
Jiaming Yuan
1877cb8e83
Change default metric for gamma regression to deviance. (#9757)
* Change default metric for gamma regression to deviance.

- Cleanup the gamma implementation.
- Use deviance instead since the objective is derived from deviance.
2023-11-22 21:17:48 +08:00
Jiaming Yuan
0715ab3c10
Use dlopen to load NCCL. (#9796)
This PR adds optional support for loading nccl with `dlopen` as an alternative of compile time linking. This is to address the size bloat issue with the PyPI binary release.
- Add CMake option to load `nccl` at runtime.
- Add an NCCL stub.

After this, `nccl` will be fetched from PyPI when using pip to install XGBoost, either by a user or by `pyproject.toml`. Others who want to link the nccl at compile time can continue to do so without any change.

At the moment, this is Linux only since we only support MNMG on Linux.
2023-11-22 19:27:31 +08:00