Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
6,817 Commits 32 Branches 59 Tags
Commit Graph

41 Commits

Author SHA1 Message Date
Jiaming Yuan
7a02facc9d
Serialize expand entry for allgather. (#9702) 2023-10-24 14:33:28 +08:00
Jiaming Yuan
8c676c889d
Remove internal use of gpu_id. (#9568) 2023-09-20 23:29:51 +08:00
Jiaming Yuan
1caa93221a
Use realloc for histogram cache and expose the cache limit. (#9455) 2023-08-10 14:05:27 +08:00
Jiaming Yuan
54029a59af
Bound the size of the histogram cache. (#9440) 
- A new histogram collection with a limit in size.
- Unify histogram building logic between hist, multi-hist, and approx.
 2023-08-08 03:21:26 +08:00
Jiaming Yuan
1332ff787f
Unify the code path between local and distributed training. (#9433) 
This removes the need for a local histogram space during distributed training, which cuts the cache size by half.
 2023-08-03 21:46:36 +08:00
Jiaming Yuan
22b0a55a04
Remove hist builder class. (#9400) 
* Remove hist build class.

* Cleanup this stateless class.

* Add comment to thread block.
 2023-07-22 10:43:12 +08:00
Jiaming Yuan
152e2fb072
Unify test helpers for creating ctx. (#9274) 2023-06-10 03:35:22 +08:00
Rong Ou
5b69534b43
Support column split in multi-target hist (#9171) 2023-05-26 16:56:05 +08:00
Jiaming Yuan
08ce495b5d
Use Booster context in DMatrix. (#8896) 
- Pass context from booster to DMatrix.
- Use context instead of integer for `n_threads`.
- Check the consistency configuration for `max_bin`.
- Test for all combinations of initialization options.
 2023-04-28 21:47:14 +08:00
Jiaming Yuan
8685556af2
Implement hist evaluator for multi-target tree. (#8908) 2023-03-15 01:42:51 +08:00
Jiaming Yuan
5ba3509dd3
Define multi expand entry. (#8895) 2023-03-13 19:31:05 +08:00
Jiaming Yuan
228a46e8ad
Support learning rate for zero-hessian objectives. (#8866) 2023-03-06 20:33:28 +08:00
Rong Ou
a65ad0bd9c
Support column split in histogram builder (#8811) 2023-02-17 22:37:01 +08:00
Jiaming Yuan
282b1729da
Specify the number of threads for parallel sort. (#8735) 
* Specify the number of threads for parallel sort.

- Pass context object into argsort.
- Replace macros with inline functions.
 2023-02-16 00:20:19 +08:00
Jiaming Yuan
3760cede0f
Consistent use of context to specify number of threads. (#8733) 
- Use context in all tests.
- Use context in R.
- Use context in C API DMatrix initialization. (0 threads is used as dft).
 2023-01-30 15:25:31 +08:00
Jiaming Yuan
e49e0998c0
Extract CPU sampling routines. (#8697) 2023-01-19 23:28:18 +08:00
Dmitry Razdoburdin
5bd849f1b5
Unify the partitioner for hist and approx. 
Co-authored-by: dmitry.razdoburdin <drazdobu@jfldaal005.jf.intel.com>
Co-authored-by: jiamingy <jm.yuan@outlook.com>
 2022-10-20 02:49:20 +08:00
Rory Mitchell
8f77677193
Use quantised gradients in gpu_hist histograms (#8246) 2022-09-26 17:35:35 +02:00
Dmitry Razdoburdin
eb7bbee2c9
Optional by-column histogram build. (#8233) 
Co-authored-by: dmitry.razdoburdin <drazdobu@jfldaal005.jf.intel.com>
 2022-09-22 05:16:13 +08:00
Jiaming Yuan
b5eb36f1af
Add max_cat_threshold to GPU and handle missing cat values. (#8212) 2022-09-07 00:57:51 +08:00
Jiaming Yuan
4a4e5c7c18
Prepare gradient index for Quantile DMatrix. (#8103) 
* Prepare gradient index for Quantile DMatrix.

- Implement push batch with adapter batch.
- Implement `GetFvalue` for prediction.
 2022-07-22 17:26:33 +08:00
Jiaming Yuan
142a208a90
Fix compiler warnings. (#8022) 
- Remove/fix unused parameters
- Remove deprecated code in rabit.
- Update dmlc-core.
 2022-06-22 21:29:10 +08:00
Jiaming Yuan
bde4f25794
Handle missing categorical value in CPU evaluator. (#7948) 2022-05-27 14:15:47 +08:00
Jiaming Yuan
18a38f7ca0
Refactor for GHistIndex. (#7923) 
* Pass sparse page as adapter, which prepares for quantile dmatrix.
* Remove old external memory code like `rbegin` and extra `Init` function.
* Simplify type dispatch.
 2022-05-23 23:04:53 +08:00
Jiaming Yuan
4fcfd9c96e
Fix and cleanup for column matrix. (#7901) 
* Fix missed type dispatching for dense columns with missing values.
* Code cleanup to reduce special cases.
* Reduce memory usage.
 2022-05-16 21:11:50 +08:00
Jiaming Yuan
1b6538b4e5
[breaking] Drop single precision histogram (#7892) 
Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>
 2022-05-13 19:54:55 +08:00
Jiaming Yuan
317d7be6ee
Always use partition based categorical splits. (#7857) 2022-05-03 22:30:32 +08:00
Jiaming Yuan
4d81c741e9
External memory support for hist (#7531) 
* Generate column matrix from gHistIndex.
* Avoid synchronization with the sparse page once the cache is written.
* Cleanups: Remove member variables/functions, change the update routine to look like approx and gpu_hist.
* Remove pruner.
 2022-03-22 00:13:20 +08:00
Jiaming Yuan
83a66b4994
Support categorical data for hist. (#7695) 
* Extract partitioner from hist.
* Implement categorical data support by passing the gradient index directly into the partitioner.
* Organize/update document.
* Remove code for negative hessian.
 2022-02-25 03:47:14 +08:00
Jiaming Yuan
6762c45494
Small cleanup to gradient index and hist. (#7668) 
* Code comments.
* Const accessor to index.
* Remove some weird variables in the `Index` class.
* Simplify the `MemStackAllocator`.
 2022-02-23 11:37:21 +08:00
Jiaming Yuan
0d0abe1845
Support optimal partitioning for GPU hist. (#7652) 
* Implement `MaxCategory` in quantile.
* Implement partition-based split for GPU evaluation.  Currently, it's based on the existing evaluation function.
* Extract an evaluator from GPU Hist to store the needed states.
* Added some CUDA stream/event utilities.
* Update document with references.
* Fixed a bug in approx evaluator where the number of data points is less than the number of categories.
 2022-02-15 03:03:12 +08:00
Jiaming Yuan
2775c2a1ab
Prepare external memory support for hist. (#7638) 
This PR prepares the GHistIndexMatrix to host the column matrix which is used by the hist tree method by accepting sparse_threshold parameter.

Some cleanups are made to ensure the correct batch param is being passed into DMatrix along with some additional tests for correctness of SimpleDMatrix.
 2022-02-10 16:58:02 +08:00
Jiaming Yuan
5817840858
Remove omp_get_max_threads in data. (#7588) 2022-01-24 02:44:07 +08:00
Jiaming Yuan
9ab73f737e
Extract Sketch Entry from hist maker. (#7503) 
* Extract Sketch Entry from hist maker.

* Add a new sketch container for sorted inputs.
* Optimize bin search.
 2021-12-18 05:36:56 +08:00
Jiaming Yuan
bf7bb575b4
Test CPU histogram with cat data. (#7465) 2021-11-27 00:43:28 +08:00
Jiaming Yuan
176110a22d
Support external memory in CPU histogram building. (#7372) 2021-11-23 01:13:33 +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
8d7c6366d7
Accept histogram cut instead gradient index in evaluation. (#7336) 2021-10-20 18:04:46 +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
149f209af6
Extract histogram builder from CPU Hist. (#7152) 
* Extract the CPU histogram builder.
* Fix tests.
* Reduce number of histograms being built.
 2021-08-09 21:15:21 +08:00
Jiaming Yuan
615ab2b03e
Extract evaluate splits from CPU hist. (#7079) 
Other than modularizing the split evaluation function, this PR also removes some more functions including `InitNewNodes` and `BuildNodeStats` among some other unused variables.  Also, scattered code like setting leaf weights is grouped into the split evaluator and `NodeEntry` is simplified and made private.  Another subtle difference with the original implementation is that the modified code doesn't call `tree[nidx].Parent()` to traversal upward.
 2021-07-07 15:16:25 +08:00