)

* added type hints to custom_obj and custom_eval for Spark persistence


Co-authored-by: Bobby Wang <wbo4958@gmail.com>

…mlc#7358)

* Support binary/multi-class classification, ranking.
* Add documents.
* Handle missing data.

We have 2 new custom objective demos covering both regression and classification with
accompanying tutorials in documents.

* Replace -1 in pandas initializer.
* Unify `IsValid` functor.
* Mimic pandas data handling in cuDF glue code.
* Check invalid categories.
* Fix DDM sketching.

* Add new classifiers.
* Typehint.

…mlc#6751)


A new parameter `custom_metric` is added to `train` and `cv` to distinguish the behaviour from the old `feval`.  And `feval` is deprecated.  The new `custom_metric` receives transformed prediction when the built-in objective is used.  This enables XGBoost to use cost functions from other libraries like scikit-learn directly without going through the definition of the link function.

`eval_metric` and `early_stopping_rounds` in sklearn interface are moved from `fit` to `__init__` and is now saved as part of the scikit-learn model.  The old ones in `fit` function are now deprecated. The new `eval_metric` in `__init__` has the same new behaviour as `custom_metric`.

Added more detailed documents for the behaviour of custom objective and metric.

Spark 3.2 depends on 3.7.0-M11 which has changed some implicited functions'
signatures. And it will result the xgboost4j built against spark 3.0/3.1
failed when saving the model.

Co-authored-by: Hyunsu Cho <chohyu01@cs.washington.edu>

Generated using `clang-format -style=google -dump-config > .clang-format`, with column
width changed from 80 to 100 to be consistent with existing cpplint check.

* Fix span reverse iterator.

* Disable `rbegin` on device code to avoid calling host function.
* Add `trbegin` and friends.

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.

* Support building with CTK11.5.

* Require system cub installation for CTK11.4+.
* Check thrust version for segmented sort.

* Move attribute setter to callback.
* Remove the internal train function.
* Remove unnecessary initialization.

* Add test for invalid categorical data values.

* Add check during sketching.

Change from system Python to environment python3.  For Ubuntu 20.04, only `python3` is
available and there's no `python`.  So at least `python3` is consistent with Python
virtual env, Ubuntu and anaconda.

* Define the `ObjInfo` and pass it down to every tree updater.

* Replace existing matrix and vector view.

This is to prepare for handling higher dimension data and prediction when we support multi-target models.

dmlc#7396)

* [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.

* Update dmlc-core.
* New parquet parser in dmlc-core.
* Use data iter for GPU sampling tests.

* Implementation for one hot based.
* Implementation for partition based. (LightGBM)

* Remove outdated workaround and description.

1. Add PreXGBoost to build RDD[Watches] from Dataset
2. Feed RDD[Watches] built from PreXGBoost to XGBoost to train

The patch is too big to be backported.

* Add equality operators.
* Return a view in substr.
* Add proper iterator types.

* Add `Tensor` class.
* Add elementwise kernel for CPU and GPU.
* Add unravel index.
* Move some computation to compile time.

* 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.

* Remove hist builder dtor.

* Initialize values.

* Tolerance.

* Remove the use of nthread in col maker.

* [CI] Drop CUDA 10.1; Require 11.0

* Change NCCL version

* Use CUDA 10.1 for clang-tidy, for now

* Remove JDK 11 and 12

* Fix NCCL version

* Don't require 11.0 just yet, until clang-tidy is fixed

* Skip MultiClassesSerializationTest.GpuHist

extract the common part of transform code from XGBoostClassifier
and XGBoostRegressor

* Add eta test.
* Don't skip test.

This PR changes base_margin into a 3-dim array, with one of them being reserved for multi-target classification. Also, a breaking change is made for binary serialization due to extra dimension along with a fix for saving the feature weights. Lastly, it unifies the prediction initialization between CPU and GPU. After this PR, the meta info setter in Python will be based on array interface.

* [CI] Add a helper script to aid Maven release

* Move script to dev/ [skip ci]

* Update command [skip ci]

* Use sphinx gallery to integrate examples.
* Remove mock objects.
* Add dask doc inventory.

* Split up the tests.

The documentation states that to build from source you need CMake 3.13 or higher. However, according to https://github.com/dmlc/xgboost/blob/master/CMakeLists.txt#L1 CMake 3.14 or higher is required.

* bost -> both

* doc improvement

* use original filename

* syntax highlight false

* missed a few highlights

* Fix empty partition.

* war.

* Add a new ctor to tensor for `initilizer_list`.
* Change labels from host device vector to tensor.
* Rename the field from `labels_` to `labels` since it's a public member.

* Add GPU pre-processing pipeline.

* Extract Sketch Entry from hist maker.

* Add a new sketch container for sorted inputs.
* Optimize bin search.

* Add num target model parameter, which is configured from input labels.
* Change elementwise metric and indexing for weights.
* Add demo.
* Add tests.

…mlc#7481)

Instead of accessing data from the `original_page_`, access the data from the first page of the available batch.

fix dmlc#7476

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

As suggested on issue dmlc#7480

occurance -> occurrence

This is part of the on going effort to remove the dependency on global omp variables.

* Add support in sklearn classifier.

* Define `feature_names_in_`.
* Raise attribute error if it's not defined.

* Use cross references instead of URL.
* Fix auto doc for callback.

* Unused variable.
* Unnecessary virtual function.

* Use ctx in gbm.

* Use ctx threads in gbm and linear.

* Add note.

* Fix n boosted rounds.