Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
5,205 Commits 32 Branches 59 Tags
55ee2bd77f3ff319ccb27d7d1c2780f938aefc34
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Jiaming Yuan 1b70a323a7 Improve string view to reduce string allocation. (#6644) 2021-01-27 19:08:52 +08:00
Jiaming Yuan bc08e0c9d1 Remove experimental_json_serialization from tests. (#6640) 2021-01-27 17:44:49 +08:00
Jiaming Yuan 8968ca7c0a Disable s390x and arm64 tests on travis for now. (#6641) 2021-01-27 16:21:40 +08:00
Jiaming Yuan d19a0ddacf Move sdist test to action. (#6635) 
* Move x86 linux and osx sdist test to action.

* Add Windows.
 2021-01-26 08:25:59 +08:00
Jiaming Yuan 740d042255 Add base_margin for evaluation dataset. (#6591) 
* Add base margin to evaluation datasets.
* Unify the code base for evaluation matrices.
 2021-01-26 02:11:02 +08:00
Jiaming Yuan 4bf23c2391 Specify shape in prediction contrib and interaction. (#6614) 2021-01-26 02:08:22 +08:00
Jiaming Yuan 8942c98054 Define metainfo and other parameters for all DMatrix interfaces. (#6601) 
This PR ensures all DMatrix types have a common interface.

* Fix logic in avoiding duplicated DMatrix in sklearn.
* Check for consistency between DMatrix types.
* Add doc for bounds.
 2021-01-25 16:06:06 +08:00
Jiaming Yuan 561809200a Fix document for tree methods. (#6633) 2021-01-25 15:52:08 +08:00
Jiaming Yuan a275f40267 [dask] Rework base margin test. (#6627) 2021-01-22 17:49:13 +08:00
Jiaming Yuan 7bc56fa0ed Use simple print in tracker print function. (#6609) 2021-01-21 21:15:43 +08:00
Jiaming Yuan 26982f9fce Skip unused CMake argument in setup.py (#6611) 2021-01-21 17:25:33 +08:00
Jiaming Yuan f0fd7629ae Add helper script and doc for releasing pip package. (#6613) 
* Fix `long_description_content_type`.
 2021-01-21 14:46:52 +08:00
Jiaming Yuan f8bb678c67 Exclude dmlc test on github action. (#6625) 2021-01-20 18:50:20 +08:00
Jiaming Yuan d6d72de339 Revert ntree limit fix (#6616) 
The old (before fix) best_ntree_limit ignores the num_class parameters, which is incorrect. In before we workarounded it in c++ layer to avoid possible breaking changes on other language bindings. But the Python interpretation stayed incorrect. The PR fixed that in Python to consider num_class, but didn't remove the old workaround, so tree calculation in predictor is incorrect, see PredictBatch in CPUPredictor.
 2021-01-19 23:51:16 +08:00
Jiaming Yuan d132933550 Remove type check for solaris. (#6610) 2021-01-16 02:58:19 +08:00
Jiaming Yuan d356b7a071 Restore unknown data support. (#6595) 2021-01-14 04:51:16 +08:00
Jiaming Yuan 89a00a5866 [dask] Random forest estimators (#6602) 2021-01-13 20:59:20 +08:00
Jiaming Yuan 0027220aa0 [breaking] Remove duplicated predict functions, Fix attributes IO. (#6593) 
* Fix attributes not being restored.
* Rename all `data` to `X`. [breaking]
 2021-01-13 16:56:49 +08:00
Jiaming Yuan 03cd087da1 Remove duplicated DMatrix. (#6592) 2021-01-12 09:36:56 +08:00
Jiaming Yuan c709f2aaaf Fix evaluation result for XGBRanker. (#6594) 
* Remove duplicated code, which fixes typo `evals_result` -> `evals_result_`.
 2021-01-12 09:36:41 +08:00
Jiaming Yuan f2f7dd87b8 Use view for SparsePage exclusively. (#6590) 2021-01-11 18:04:55 +08:00
Jiaming Yuan 78f2cd83d7 Suppress hypothesis health check for dask client. (#6589) 2021-01-11 14:11:57 +08:00
Jiaming Yuan 80065d571e [dask] Add DaskXGBRanker (#6576) 
* Initial support for distributed LTR using dask.

* Support `qid` in libxgboost.
* Refactor `predict` and `n_features_in_`, `best_[score/iteration/ntree_limit]`
  to avoid duplicated code.
* Define `DaskXGBRanker`.

The dask ranker doesn't support group structure, instead it uses query id and
convert to group ptr internally.
 2021-01-08 18:35:09 +08:00
Jiaming Yuan 96d3d32265 [dask] Add shap tests. (#6575) 2021-01-08 14:59:27 +08:00
Jiaming Yuan 7c9dcbedbc Fix best_ntree_limit for dart and gblinear. (#6579) 2021-01-08 10:05:39 +08:00
Jiaming Yuan f5ff90cd87 Support _estimator_type. (#6582) 
* Use `_estimator_type`.

For more info, see: https://scikit-learn.org/stable/developers/develop.html#estimator-types

* Model trained from dask can be loaded by single node skl interface.
 2021-01-08 10:01:16 +08:00
Jiaming YuanandHyunsu Cho 8747885a8b Support Solaris. (#6578) 
* Add system header.

* Remove use of TR1 on Solaris

Co-authored-by: Hyunsu Cho <chohyu01@cs.washington.edu>
 2021-01-07 09:05:05 +08:00
Jiaming Yuan 60cfd14349 [dask, sklearn] Fix predict proba. (#6566) 
* For sklearn:
  - Handles user defined objective function.
  - Handles `softmax`.

* For dask:
  - Use the implementation from sklearn, the previous implementation doesn't perform any extra handling.
 2021-01-05 08:29:06 +08:00
Jiaming Yuan 516a93d25c Fix best_ntree_limit. (#6569) 2021-01-03 05:58:54 +08:00
Jiaming Yuan 2b049b32e9 Document various tree methods. (#6564) 2021-01-02 15:40:46 +08:00
Jiaming Yuan 5e9e525223 Remove warnings in tests. (#6554) 2020-12-31 13:41:18 +08:00
Jiaming YuanandHyunsu Cho de8fd852a5 [dask] Add type hints. (#6519) 
* Add validate_features.
* Show type hints in doc.

Co-authored-by: Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-12-29 19:41:02 +08:00
Jiaming Yuan 610ee632cc [Breaking] Rename data to X in predict_proba. (#6555) 
New Scikit-Learn version uses keyword argument, and `X` is the predefined
keyword.

* Use pip to install latest Python graphviz on Windows CI.
 2020-12-28 21:36:03 +08:00
Jiaming Yuan cb207a355d Add script for generating release tarball. (#6544) 2020-12-23 16:08:10 +08:00
Jiaming YuanandPhilip Hyunsu Cho ca3da55de4 Support early stopping with training continuation, correct num boosted rounds. (#6506) 
* Implement early stopping with training continuation.

* Add new C API for obtaining boosted rounds.

* Fix off by 1 in `save_best`.

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-12-17 19:59:19 +08:00
Jiaming Yuan d8d684538c [CI] Split up main.yml, add mypy. (#6515) 2020-12-17 00:15:44 +08:00
Jiaming Yuan c5876277a8 Drop saving binary format for memory snapshot. (#6513) 2020-12-17 00:14:57 +08:00
Jiaming Yuan 0e97d97d50 Fix merge conflict. (#6512) 2020-12-16 18:02:25 +08:00
Jiaming Yuan 347f593169 Accept numpy array for DMatrix slice index. (#6368) 2020-12-16 14:42:52 +08:00
Jiaming YuanandPhilip Hyunsu Cho ef4a0e0aac Fix DMatrix feature names/types IO. (#6507) 
* Fix feature names/types IO

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-12-16 14:24:27 +08:00
Jiaming Yuan 886486a519 Support categorical data in GPU weighted sketching. (#6508) 2020-12-16 14:23:28 +08:00
Jiaming Yuan 3c3f026ec1 Move metric configuration into booster. (#6504) 2020-12-16 05:35:04 +08:00
Jiaming Yuan d45c0d843b Show partition status in dask error. (#6366) 2020-12-16 02:58:21 +08:00
Jiaming Yuan a30461cf87 [dask] Support all parameters in regressor and classifier. (#6471) 
* Add eval_metric.
* Add callback.
* Add feature weights.
* Add custom objective.
 2020-12-14 07:35:56 +08:00
Jiaming Yuan 0ffaf0f5be Fix dask ip resolution. (#6475) 
This adopts the solution used in dask/dask-xgboost#40 which employs the get_host_ip from dmlc-core tracker.
 2020-12-07 16:36:23 -08:00
Jiaming Yuan 47b86180f6 Don't validate feature when number of rows is 0. (#6472) 2020-12-07 18:08:51 +08:00
Jiaming Yuan 703c2d06aa Fix global config default value. (#6470) 2020-12-06 06:15:33 +08:00
Jiaming YuanandHyunsu Cho d6386e45e8 Fix filtering callable objects in skl xgb param. (#6466) 
Co-authored-by: Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-12-05 17:20:36 +08:00
Jiaming Yuan 927c316aeb Fix period in evaluation monitor. (#6441) 2020-11-29 03:18:33 +08:00
Jiaming Yuan f4ff1c53fd Fix CLI ranking demo. (#6439) 
Save model at final round.
 2020-11-29 03:12:06 +08:00
Jiaming Yuan 42d31d9dcb Fix MPI build. (#6403) 2020-11-21 13:38:21 +08:00
Jiaming Yuan 2ce2a1a4d8 [SKL] Propagate parameters to booster during set_param. (#6416) 2020-11-20 20:37:35 +08:00
Jiaming Yuan 00218d065a [dask] Update document. [skip ci] (#6413) 2020-11-20 19:16:19 +08:00
Jiaming Yuan c120822a24 Fix flaky sparse page dmatrix test. (#6417) 2020-11-20 19:15:45 +08:00
Jiaming Yuan a7b42adb74 Fix dask predict (#6412) 2020-11-20 10:10:52 +08:00
Jiaming Yuan 44a9d69efb Small cleanup to evaluator. (#6400) 2020-11-20 09:33:51 +08:00
Jiaming Yuan 3ac173fc8b Fix typo. (#6399) 2020-11-16 16:59:12 -08:00
Jiaming Yuan fcd6fad822 [dask] Small cleanup. (#6391) 2020-11-14 22:15:05 +08:00
Jiaming Yuan 4ccf92ea34 [dask] Fix union of workers. (#6375) 2020-11-13 16:55:05 +08:00
Jiaming Yuan fcfeb4959c Deprecate positional arguments. (#6365) 
Deprecate positional arguments in following functions:

- `__init__` for all classes in sklearn module.
- `fit` method for all classes in sklearn module.
- dask interface.
- `set_info` for `DMatrix` class.

Refactor the evaluation matrices handling.
 2020-11-13 11:10:30 +08:00
Jiaming Yuan c1a62b5fa2 Expect gpu external memory to fail. (#6381) 2020-11-12 19:24:48 +08:00
Jiaming Yuan c90f968d92 Update Python documents. (#6376) 2020-11-12 17:51:32 +08:00
Jiaming Yuan d711d648cb Fix label errors in graph visualization (#6369) 2020-11-11 17:44:59 -08:00
Jiaming Yuan debeae2509 [R] Fix warnings from R check --as-cran (#6374) 
* Remove exit and printf.

* Fix warnings.
 2020-11-11 18:39:37 +08:00
Jiaming Yuan 6e12c2a6f8 [dask] Supoort running on GKE. (#6343) 
* Avoid accessing `scheduler_info()['workers']`.
* Avoid calling `client.gather` inside task.
* Avoid using `client.scheduler_address`.
 2020-11-11 18:04:34 +08:00
Jiaming Yuan 8a17610666 Implement GPU predict leaf. (#6187) 2020-11-11 17:33:47 +08:00
Jiaming YuanandPhilip Hyunsu Cho a5cfa7841e Run R check as cran on action. [skip ci] (#6371) 
Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-11-11 12:02:53 +08:00
Jiaming Yuan 43efadea2e Deterministic data partitioning for external memory (#6317) 
* Make external memory data partitioning deterministic.

* Change the meaning of `page_size` from bytes to number of rows.

* Design a data pool.

* Note for external memory.

* Enable unity build on Windows CI.

* Force garbage collect on test.
 2020-11-11 06:11:06 +08:00
Jiaming Yuan e65e3cf36e Support shared library in system path. (#6362) 2020-11-10 16:04:25 +08:00
Jiaming Yuan 184e2eac7d Add period to evaluation monitor. (#6348) 2020-11-10 07:47:48 +08:00
Jiaming Yuan 519cee115a Avoid resetting seed for every configuration. (#6349) 2020-11-06 10:28:35 +08:00
Jiaming YuanandPhilip Hyunsu Cho 2cc9662005 Support slicing tree model (#6302) 
This PR is meant the end the confusion around best_ntree_limit and unify model slicing. We have multi-class and random forests, asking users to understand how to set ntree_limit is difficult and error prone.

* Implement the save_best option in early stopping.

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-11-02 23:27:39 -08:00
Jiaming Yuan 7756192906 [dask] Fix prediction on DaskDMatrix with multiple meta data. (#6333) 
* Unify the meta handling methods.
 2020-11-02 19:18:44 -05:00
Jiaming Yuan 5a7b3592ed Optional find_package for sanitizers. (#6329) 2020-11-02 19:17:17 -05:00
Jiaming Yuan 048acf81cd Enable shap sparse test. (#6332) 2020-11-01 20:59:27 +08:00
Jiaming Yuan 6ff331b705 Fix Python callback. (#6320) 2020-10-30 05:03:44 +08:00
Jiaming Yuan c80657b542 Fix flaky data initialization test. (#6318) 2020-10-30 03:11:22 +08:00
Jiaming Yuan 74ea82209b Lazy import dask libraries. (#6309) 
* Lazy import dask libraries.

* Lint && fix.

* Use short name.
 2020-10-28 15:50:11 -07:00
Jiaming Yuan dfac5f89e9 Group CLI demo into subdirectory. (#6258) 
CLI is not most developed interface. Putting them into correct directory can help new users to avoid it as most of the use cases are from a language binding.
 2020-10-28 14:40:44 -07:00
Jiaming Yuan e8884c4637 Document tree method for feature weights. (#6312) 2020-10-28 13:42:13 -07:00
Jiaming Yuan c4da967b5c Support unity build. (#6295) 
* Support unity build.

* Setup on Windows Jenkins.

* Revert "Setup on Windows Jenkins."

This reverts commit 8345cb8d2b009eec8ae9fa6f16412a7c9b6ec12c.
 2020-10-28 11:49:28 -07:00
Jiaming Yuan 3310e208fd Fix inplace prediction interval. (#6259) 
* Add back the interval in call.
* Make the interval non-optional.
 2020-10-28 13:13:59 +08:00
Jiaming Yuan cc76724762 Reduce warning. (#6273) 2020-10-27 12:24:19 -07:00
Jiaming Yuan b180223d18 Cleanup RABIT. (#6290) 
* Remove recovery and MPI speed tests.
* Remove readme.
* Remove Python binding.
* Add checks in C API.
 2020-10-27 08:48:22 +08:00
Jiaming Yuan bcfab4d726 Revert "Disable JSON full serialization for now. (#6248)" (#6266) 
This reverts commit 6d293020fb.
 2020-10-27 03:30:47 +08:00
Jiaming Yuan d61b628bf5 Remove RABIT CMake targets. (#6275) 
* Now it's built as part of libxgboost.
* Set correct C API error in RABIT initialization and finalization.
* Remove redundant message.
* Guard the tracker print C API.
 2020-10-27 01:30:20 +08:00
Jiaming YuanandHyunsu Cho 2686d32a36 Skip dask tests on ARM. (#6267) 
Co-authored-by: Hyunsu Cho <chohyu01@cs.washington.edu>
 2020-10-26 15:09:05 +08:00
Jiaming Yuan b5c2a47b20 Drop single point model recovery (#6262) 
* Pass rabit params in JVM package.
* Implement timeout using poll timeout parameter.
* Remove OOB data check.
 2020-10-21 15:27:03 +08:00
Jiaming Yuan 81c37c28d5 Time the CPU tests on Jenkins. (#6257) 
* Time the CPU tests on Jenkins.
* Reduce thread contention.
* Add doc.
* Skip heavy tests on ARM.
 2020-10-20 17:19:07 -07:00
Jiaming Yuan ddf37cca30 Unify thread configuration. (#6186) 2020-10-19 16:05:42 +08:00
Jiaming Yuan 5037abeb86 Fix linear gpu input (#6255) 2020-10-19 12:02:36 +08:00
Jiaming Yuan 6d293020fb Disable JSON full serialization for now. (#6248) 
* Disable JSON serialization for now.

* Multi-class classification is checkpointing for each iteration.
This brings significant overhead.

Revert: 90355b4f00

* Set R tests to use binary.
 2020-10-16 17:59:54 +08:00
Jiaming Yuan 52452bebb9 Fix cls typo. (#6247) 2020-10-16 16:40:44 +08:00
Jiaming Yuan 3da5a69dc9 Fix typo in dask interface. (#6240) 2020-10-15 15:26:29 +08:00
Jiaming Yuan bed7ae4083 Loop over thrust::reduce. (#6229) 
* Check input chunk size of dqdm.
* Add doc for current limitation.
 2020-10-14 10:40:56 +13:00
Jiaming Yuan b05073bda5 [dask] Test for data initializaton. (#6226) 2020-10-13 11:08:35 +08:00
Jiaming Yuan 2443275891 Cleanup Python code. (#6223) 
* Remove pathlike as XGBoost 1.2 requires Python 3.6.
* Move conditional import of dask/distributed into dask module.
 2020-10-12 15:44:41 +08:00
Jiaming Yuan 70c2039748 Catch all standard exceptions in C API. (#6220) 
* `std::bad_alloc` is not guaranteed to be caught.
 2020-10-12 14:01:46 +08:00
Jiaming Yuan 2241563f23 Handle duplicated values in sketching. (#6178) 
* Accumulate weights in duplicated values.
* Fix device id in iterative dmatrix.
 2020-10-10 19:32:44 +08:00
Jiaming Yuan ab5b35134f Rework Python callback functions. (#6199) 
* Define a new callback interface for Python.
* Deprecate the old callbacks.
* Enable early stopping on dask.
 2020-10-10 17:52:36 +08:00