Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
6,011 Commits 32 Branches 59 Tags
cfd2a9f8724ada8dbae312019bb0ab548e1c9b13
Commit Graph
1328 Commits
Author SHA1 Message Date
Jiaming Yuan 55ee272ea8 Extend array interface to handle ndarray. (#7434) 
* Extend array interface to handle ndarray.

The `ArrayInterface` class is extended to support multi-dim array inputs. Previously this
class handles only 2-dim (vector is also matrix).  This PR specifies the expected
dimension at compile-time and the array interface can perform various checks automatically
for input data. Also, adapters like CSR are more rigorous about their input.  Lastly, row
vector and column vector are handled without intervention from the caller.
 2021-11-16 09:52:15 +08:00
Jiaming Yuan d4274bc556 Fix typo. (#7433) 2021-11-15 01:28:11 +08:00
Jiaming Yuan a7057fa64c Implement typed storage for tensor. (#7429) 
* Add `Tensor` class.
* Add elementwise kernel for CPU and GPU.
* Add unravel index.
* Move some computation to compile time.
 2021-11-14 18:53:13 +08:00
Jiaming Yuan 46726ec176 Expose build info (#7399) 2021-11-12 18:22:46 +08:00
Jiaming Yuan 937fa282b5 Extract string view. (#7416) 
* Add equality operators.
* Return a view in substr.
* Add proper iterator types.
 2021-11-12 18:22:30 +08:00
Jiaming Yuan ca6f980932 Check number of trees in inplace predict. (#7409) 2021-11-12 18:20:23 +08:00
Jiaming Yuan d7d1b6e3a6 CPU evaluation for cat data. (#7393) 
* Implementation for one hot based.
* Implementation for partition based. (LightGBM)
 2021-11-06 14:41:35 +08:00
Jiaming Yuan 6ede12412c Update dmlc-core and use data iter for GPU sampling tests. (#7398) 
* Update dmlc-core.
* New parquet parser in dmlc-core.
* Use data iter for GPU sampling tests.
 2021-11-06 05:12:49 +08:00
Jiaming Yuan c968217ca8 [R] Fix global feature importance and predict with 1 sample. (#7394) 
* [R] Fix global feature importance.

* Add implementation for tree index.  The parameter is not documented in C API since we
should work on porting the model slicing to R instead of supporting more use of tree
index.

* Fix the difference between "gain" and "total_gain".

* debug.

* Fix prediction.
 2021-11-05 10:07:00 +08:00
Jiaming Yuan b06040b6d0 Implement a general array view. (#7365) 
* Replace existing matrix and vector view.

This is to prepare for handling higher dimension data and prediction when we support multi-target models.
 2021-11-05 04:16:11 +08:00
Jiaming Yuan 4100827971 Pass infomation about objective to tree methods. (#7385) 
* Define the `ObjInfo` and pass it down to every tree updater.
 2021-11-04 01:52:44 +08:00
Jiaming Yuan ccdabe4512 Support building gradient index with cat data. (#7371) 2021-11-03 22:37:37 +08:00
Jiaming Yuan 57a4b4ff64 Handle OMP_THREAD_LIMIT. (#7390) 2021-11-03 15:44:38 +08:00
Jiaming Yuan a55d43ccfd Add test for invalid categorical data values. (#7380) 
* Add test for invalid categorical data values.

* Add check during sketching.
 2021-11-02 18:00:52 +08:00
Jiaming Yuan 32e673d8c4 Support building with CTK11.5. (#7379) 
* Support building with CTK11.5.

* Require system cub installation for CTK11.4+.
* Check thrust version for segmented sort.
 2021-11-02 16:22:26 +08:00
Jiaming Yuan a13321148a Support multi-class with base margin. (#7381) 
This is already partially supported but never properly tested. So the only possible way to use it is calling `numpy.ndarray.flatten` with `base_margin` before passing it into XGBoost. This PR adds proper support
for most of the data types along with tests.
 2021-11-02 13:38:00 +08:00
Jiaming Yuan 6295dc3b67 Fix span reverse iterator. (#7387) 
* Fix span reverse iterator.

* Disable `rbegin` on device code to avoid calling host function.
* Add `trbegin` and friends.
 2021-11-02 13:35:59 +08:00
Jiaming Yuan 0f7a9b42f1 Use double precision in metric calculation. (#7364) 2021-11-02 12:00:32 +08:00
Jiaming Yuan d05754f558 Avoid OMP reduction in AUC. (#7362) 2021-10-28 05:03:52 +08:00
Jiaming Yuan ac9bfaa4f2 Handle missing values in dataframe with category dtype. (#7331) 
* Replace -1 in pandas initializer.
* Unify `IsValid` functor.
* Mimic pandas data handling in cuDF glue code.
* Check invalid categories.
* Fix DDM sketching.
 2021-10-28 03:33:54 +08:00
Jiaming Yuan d4349426d8 Re-implement PR-AUC. (#7297) 
* Support binary/multi-class classification, ranking.
* Add documents.
* Handle missing data.
 2021-10-26 13:07:50 +08:00
Jiaming Yuan fd61c61071 Avoid omp reduction in rank metric. (#7349) 2021-10-22 14:13:34 +08:00
Jiaming Yuan d1f00fb0b7 Stricter validation for group. (#7345) 2021-10-21 12:13:33 +08:00
Jiaming Yuan 8d7c6366d7 Accept histogram cut instead gradient index in evaluation. (#7336) 2021-10-20 18:04:46 +08:00
Jiaming Yuan fbb0dc4275 Remove auto configuration of seed_per_iteration. (#7009) 
* Remove auto configuration of seed_per_iteration.

This should be related to model recovery from rabit, which is removed.

* Document.
 2021-10-17 15:58:57 +08:00
Jiaming Yuan fb1a9e6bc5 Avoid omp reduction in coordinate descent and aft metrics. (#7316) 
Aside from the omp issue, parameter configuration for aft metric is simplified.
 2021-10-17 15:55:49 +08:00
Jiaming Yuan 8e619010d0 Extract CPUExpandEntry and HistParam. (#7321) 
* Remove kRootNid.
* Check for empty hessian.
 2021-10-17 14:22:25 +08:00
Jiaming Yuan 4ddf8d001c Deterministic result for element-wise/mclass metrics. (#7303) 
Remove openmp reduction.
 2021-10-13 14:22:40 +08:00
Jiaming Yuan a7d0c66457 Remove unused code. (#7293) 2021-10-12 15:04:41 +08:00
Jiaming Yuan 298af6f409 Fix weighted samples in multi-class AUC. (#7300) 2021-10-11 15:12:29 +08:00
Jiaming Yuan d8cb395380 Fix gamma neg log likelihood. (#7275) 2021-10-05 16:57:08 +08:00
Jiaming Yuan b3b03200e2 Remove old warning in 1.3 (#7279) 2021-10-01 08:05:50 +08:00
Jiaming Yuan d8a549e6ac Avoid thread block with sparse data. (#7255) 2021-09-25 13:11:34 +08:00
Jiaming YuanandPhilip Hyunsu Cho ca17f8a5fc Dispatch thrust versions and upgrade rmm. (#7254) 
Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2021-09-25 03:43:23 +08:00
ShvetsKSandfis 475fd1abec Reduced span overheads in objective function calculate (#7206) 
Co-authored-by: fis <jm.yuan@outlook.com>
 2021-09-23 04:43:59 +08:00
david-cortes 4f93e5586a Improve wording for warning (#7248) 
This warning sounds  a bit ungrammatical. Additionally, the second part of the warning is not clear. This PR changes the wording to make it clearer.
 2021-09-21 10:48:11 +08:00
Jiaming Yuan c311a8c1d8 Enable compiling with system cub. (#7232) 
- Tested with all CUDA 11.x.
- Workaround cub scan by using discard iterator in AUC.
- Limit the size of Argsort when compiled with CUDA cub.
 2021-09-17 14:28:18 +08:00
Jiaming Yuan 22d56cebf1 Encode pandas categorical data automatically. (#7231) 2021-09-17 11:09:55 +08:00
Jiaming Yuan 31c1e13f90 Categorical data support in CPU sketching. (#7221) 2021-09-17 04:37:09 +08:00
Jiaming Yuan 0ed979b096 Support more input types for categorical data. (#7220) 
* Support more input types for categorical data.

* Shorten the type name from "categorical" to "c".
* Tests for np/cp array and scipy csr/csc/coo.
* Specify the type for feature info.
 2021-09-16 20:39:30 +08:00
Jiaming Yuan 2942dc68e4 Fix mixed types in GPU sketching. (#7228) 2021-09-16 00:10:25 +08:00
Jiaming Yuan 3515931305 Initial support for external memory in gradient index. (#7183) 
* Add hessian to batch param in preparation of new approx impl.
* Extract a push method for gradient index matrix.
* Use span instead of vector ref for hessian in sketching.
* Create a binary format for gradient index.
 2021-09-13 12:40:56 +08:00
Jiaming Yuan 804b2ac60f Expose DMatrix API for CUDA columnar and array. (#7217) 
* Use JSON encoded configurations.
* Expose them into header file.
 2021-09-09 17:55:25 +08:00
Jiaming Yuan b12e7f7edd Add noexcept to JSON objects. (#7205) 2021-09-07 13:56:48 +08:00
Jiaming Yuan 3a4f51f39f Avoid calling CUDA code on CPU for linear model. (#7154) 2021-09-01 10:45:31 +08:00
Jiaming Yuan ba69244a94 Restore the custom double atomic add. (#7198) 2021-08-28 18:30:42 +08:00
Jiaming YuanandAndy Adinets 7a1d67f9cb [breaking] Use integer atomic for GPU histogram. (#7180) 
On GPU we use rouding factor to truncate the gradient for deterministic results. This PR changes the gradient representation to fixed point number with exponent aligned with rounding factor.

    [breaking] Drop non-deterministic histogram.
    Use fixed point for shared memory.

This PR is to improve the performance of GPU Hist. 

Co-authored-by: Andy Adinets <aadinets@nvidia.com>
 2021-08-28 05:17:05 +08:00
Jiaming Yuan e7d7ab6bc3 Better error message for ncclUnhandledCudaError. (#7190) 2021-08-27 10:29:22 +08:00
Jiaming Yuan ee8d1f5ed8 Fix histogram truncation. (#7181) 
* Fix truncation.

* Lint.

* lint.
 2021-08-24 18:34:32 -07:00
Jiaming Yuan bf562bd33c Remove unused code. (#7175) 2021-08-18 14:02:19 +08:00