All Versions
Latest Version
Avg Release Cycle
89 days
Latest Release
570 days ago

Changelog History
Page 2

  • v1.0.2 Changes

    March 04, 2020

    πŸš€ This patch release applies the following patches to 1.0.0 release:

    • πŸ›  Fix a small typo in that broke multiple eval metrics (#5341)
    • βͺ Restore loading model from buffer. (#5360)
    • πŸ‘‰ Use type name for data type check. (#5364)
  • v1.0.1 Changes

    February 21, 2020

    πŸš€ This release is identical to the 1.0.0 release, except that it fixes a small bug that rendered 1.0.0 incompatible with Python 3.5. See #5328.

  • v1.0.0 Changes

    February 20, 2020

    v1.0.0 (2020.02.19)

    πŸš€ This release marks a major milestone for the XGBoost project.

    πŸ’… Apache-style governance, contribution policy, and semantic versioning (#4646, #4659)

    πŸ‘ Better performance scaling for multi-core CPUs (#4502, #4529, #4716, #4851, #5008, #5107, #5138, #5156)

    • πŸš€ Poor performance scaling of the hist algorithm for multi-core CPUs has been under investigation (#3810). Previous effort #4529 was replaced with a series of pull requests (#5107, #5138, #5156) aimed at achieving the same performance benefits while keeping the C++ codebase legible. The latest performance benchmark results show up to 5x speedup on Intel CPUs with many cores. Note: #5244, which concludes the effort, will become part of the upcoming release 1.1.0.

    πŸ‘Œ Improved installation experience on Mac OSX (#4672, #5074, #5080, #5146, #5240)

    • 0️⃣ It used to be quite complicated to install XGBoost on Mac OSX. XGBoost uses OpenMP to distribute work among multiple CPU cores, and Mac's default C++ compiler (Apple Clang) does not come with OpenMP. Existing work-around (using another C++ compiler) was complex and prone to fail with cryptic diagnosis (#4933, #4949, #4969).
    • Now it only takes two commands to install XGBoost: brew install libomp followed by pip install xgboost. The installed XGBoost will use all CPU cores.
    • πŸ‘€ Even better, XGBoost is now available from Homebrew: brew install xgboost. See Homebrew/homebrew-core#50467.
    • πŸš€ Previously, if you installed the XGBoost R package using the command install.packages('xgboost'), it could only use a single CPU core and you would experience slow training performance. With 1.0.0 release, the R package will use all CPU cores out of box.

    Distributed XGBoost now available on Kubernetes (#4621, #4939)

    πŸ’Ž Ruby binding for XGBoost (#4856)

    πŸ†• New Native Dask interface for multi-GPU and multi-node scaling (#4473, #4507, #4617, #4819, #4907, #4914, #4941, #4942, #4951, #4973, #5048, #5077, #5144, #5270)

    • πŸ‘€ XGBoost now integrates seamlessly with Dask, a lightweight distributed framework for data processing. Together with the first-class support for cuDF data frames (see below), it is now easier than ever to create end-to-end data pipeline running on one or more NVIDIA GPUs.
    • πŸš€ Multi-GPU training with Dask is now up to 20% faster than the previous release (#4914, #4951).

    πŸ‘ First-class support for cuDF data frames and cuPy arrays (#4737, #4745, #4794, #4850, #4891, #4902, #4918, #4927, #4928, #5053, #5189, #5194, #5206, #5219, #5225)

    • cuDF is a data frame library for loading and processing tabular data on NVIDIA GPUs. It provides a Pandas-like API.
    • cuPy implements a NumPy-compatible multi-dimensional array on NVIDIA GPUs.
    • Now users can keep the data on the GPU memory throughout the end-to-end data pipeline, obviating the need for copying data between the main memory and GPU memory.

    Feature interaction constraint is now available with approx and gpu_hist algorithms (#4534, #4587, #4596, #5034).

    Learning to rank is now GPU accelerated (#4873, #5004, #5129)

    Enable gamma parameter for GPU training (#4874, #4953)

    • The gamma parameter specifies the minimum loss reduction required to add a new split in a tree. A larger value for gamma has the effect of pre-pruning the tree, by making harder to add splits.

    External memory for GPU training (#4486, #4526, #4747, #4833, #4879, #5014)

    • πŸš€ It is now possible to use NVIDIA GPUs even when the size of training data exceeds the available GPU memory. Note that the external memory support for GPU is still experimental. #5093 will further improve performance and will become part of the upcoming release 1.1.0.
    • RFC for enabling external memory with GPU algorithms: #4357

    πŸ‘Œ Improve Scikit-Learn interface (#4558, #4842, #4929, #5049, #5151, #5130, #5227)

    • πŸš€ Many users of XGBoost enjoy the convenience and breadth of Scikit-Learn ecosystem. In this release, we revise the Scikit-Learn API of XGBoost (XGBRegressor, XGBClassifier, and XGBRanker) to achieve feature parity with the traditional XGBoost interface (xgboost.train()).
    • Insert check to validate data shapes.
    • πŸ‘ Produce an error message if eval_set is not a tuple. An error message is better than silently crashing.
    • πŸ‘ Allow using numpy.RandomState object.
    • βž• Add n_jobs as an alias of nthread.
    • Roadmap: #5152

    XGBoost4J-Spark: Redesigning checkpointing mechanism

    • RFC is available at #4786
    • πŸ‘· Clean up checkpoint file after a successful training job (#4754): The current implementation in XGBoost4J-Spark does not clean up the checkpoint file after a successful training job. If the user runs another job with the same checkpointing directory, she will get a wrong model because the second job will re-use the checkpoint file left over from the first job. To prevent this scenario, we propose to always clean up the checkpoint file after every successful training job.
    • πŸ— Avoid Multiple Jobs for Checkpointing (#5082): The current method for checkpoint is to collect the booster produced at the last iteration of each checkpoint internal to Driver and persist it in HDFS. The major issue with this approach is that it needs to re-perform the data preparation for training if the user did not choose to cache the training dataset. To avoid re-performing data prep, we build external-memory checkpointing in the XGBoost4J layer as well.
    • πŸš€ Enable deterministic repartitioning when checkpoint is enabled (#4807): Distributed algorithm for gradient boosting assumes a fixed partition of the training data between multiple iterations. In previous versions, there was no guarantee that data partition would stay the same, especially when a worker goes down and some data had to recovered from previous checkpoint. In this release, we make data partition deterministic by using the data hash value of each data row in computing the partition.

    XGBoost4J-Spark: handle errors thrown by the native code (#4560)

    • All core logic of XGBoost is written in C++, so XGBoost4J-Spark internally uses the C++ code via Java Native Interface (JNI). #4560 adds a proper error handling for any errors or exceptions arising from the C++ code, so that the XGBoost Spark application can be torn down in an orderly fashion.

    XGBoost4J-Spark: Refine method to count the number of alive cores (#4858)

    • The SparkParallelismTracker class ensures that sufficient number of executore cores are alive. To that end, it is important to query the number of alive cores reliably.

    XGBoost4J: Add BigDenseMatrix to store more than Integer.MAX_VALUE elements (#4383)

    Robust model serialization with JSON (#4632, #4708, #4739, #4868, #4936, #4945, #4974, #5086, #5087, #5089, #5091, #5094, #5110, #5111, #5112, #5120, #5137, #5218, #5222, #5236, #5245, #5248, #5281)

    In this release, we introduce an experimental support of using JSON for serializing (saving/loading) XGBoost models and related hyperparameters for training. We would like to eventually replace the old binary format with JSON, since it is an open format and parsers are available in many programming languages and platforms. See the documentation for model I/O using JSON. #3980 explains why JSON was chosen over other alternatives.

    To maximize interoperability and compatibility of the serialized models, we now split serialization into two parts (#4855):

    1. Model, e.g. decision trees and strictly related metadata like num_features.
    2. Internal configuration, consisting of training parameters and other configurable parameters. For example, max_delta_step, tree_method, objective, predictor, gpu_id.

    πŸ”§ Previously, users often ran into issues where the model file produced by one machine could not load or run on another machine. For example, models trained using a machine with an NVIDIA GPU could not run on another machine without a GPU (#5291, #5234). The reason is that the old binary format saved some internal configuration that were not universally applicable to all machines, e.g. predictor='gpu_predictor'.

    πŸ”§ Now, model saving function (Booster.save_model() in Python) will save only the model, without internal configuration. This will guarantee that your model file would be used anywhere. Internal configuration will be serialized in limited circumstances such as:

    • Multiple nodes in a distributed system exchange model details over the network.
    • Model checkpointing, to recover from possible crashes.

    πŸ‘€ This work proved to be useful for parameter validation as well (see below).

    πŸš€ Starting with 1.0.0 release, we will use semantic versioning to indicate whether the model produced by one version of XGBoost would be compatible with another version of XGBoost. Any change in the major version indicates a breaking change in the serialization format.

    πŸš€ We now provide a robust method to save and load scikit-learn related attributes (#5245). Previously, we used Python pickle to save Python attributes related to XGBClassifier, XGBRegressor, and XGBRanker objects. The attributes are necessary to properly interact with scikit-learn. See #4639 for more details. The use of pickling hampered interoperability, as a pickle from one machine may not necessarily work on another machine. Starting with this release, we use an alternative method to serialize the scikit-learn related attributes. The use of Python pickle is now discouraged (#5236, #5281).

    Parameter validation: detection of unused or incorrect parameters (#4553, #4577, #4738, #4801, #4961, #5101, #5157, #5167, #5256)

    • πŸš€ Mis-spelled training parameter is a common user mistake. In previous versions of XGBoost, mis-spelled parameters were silently ignored. Starting with 1.0.0 release, XGBoost will produce a warning message if there is any unused training parameters.
    • πŸ”§ Configuration steps now have well-defined semantics (#4542, #4738), so we know exactly where and how the internal configurable parameters are changed.
    • 🐎 The user can now use get_params() function to inspect all training parameters. This is helpful for debugging model performance.

    πŸ‘ Allow individual workers to recover from faults (#4808, #4966)

    • ⏱ Status quo: if a worker fails, all workers are shut down and restarted, and learning resumes from the last checkpoint. This involves requesting resources from the scheduler (e.g. Spark) and shuffling all the data again from scratch. Both of these operations can be quite costly and block training for extended periods of time, especially if the training data is big and the number of worker nodes is in the hundreds.
    • πŸ‘· The proposed solution is to recover the single node that failed, instead of shutting down all workers. The rest of the clusters wait until the single failed worker is bootstrapped and catches up with the rest.
    • 🚧 See roadmap at #4753. Note that this is work in progress. In particular, the feature is not yet available from XGBoost4J-Spark.

    Accurate prediction for DART models

    • πŸ‘‰ Use DART tree weights when computing SHAPs (#5050)
    • 0️⃣ Don't drop trees during DART prediction by default (#5115)
    • πŸ›  Fix DART prediction in R (#5204)

    πŸ‘‰ Make external memory more robust

    • πŸ›  Fix issues with training with external memory on cpu (#4487)
    • πŸ›  Fix crash with approx tree method on cpu (#4510)
    • πŸ›  Fix external memory race in exact (#4980). Note: dmlc::ThreadedIter is not actually thread-safe. We would like to re-design it in the long term.

    πŸ”¨ Major refactoring of the DMatrix class (#4686, #4744, #4748, #5044, #5092, #5108, #5188, #5198)

    • 🐎 Goal 1: improve performance and reduce memory consumption. Right now, if the user trains a model with a NumPy array as training data, the array gets copies 2-3 times before training begins. We'd like to reduce duplication of the data matrix.
    • Goal 2: Expose a common interface to external data, unify the way DMatrix objects are constructed and simplify the process of adding new external data sources. This work is essential for ingesting cuPy arrays.
    • RFC: #4354, Roadmap: #5143
    • πŸ‘ This work is also relevant to external memory support on GPUs.

    πŸ’₯ Breaking: XGBoost Python package now requires Python 3.5 or newer (#5021, #5274)

    • Python 3.4 has reached its end-of-life on March 16, 2019, so we now require Python 3.5 or newer.

    πŸ’₯ Breaking: GPU algorithm now requires CUDA 9.0 and higher (#4527, #4580)

    πŸ’₯ Breaking: n_gpus parameter removed; multi-GPU training now requires a distributed framework (#4579, #4749, #4773, #4810, #4867, #4908)

    • 🚚 #4531 proposed removing support for single-process multi-GPU training. Contributors would focus on multi-GPU support through distributed frameworks such as Dask and Spark, where the framework would be expected to assign a worker process for each GPU independently. By delegating GPU management and data movement to the distributed framework, we can greatly simplify the core XGBoost codebase, make multi-GPU training more robust, and reduce burden for future development.

    πŸ’₯ Breaking: Some deprecated features have been removed

    • gpu_exact training method (#4527, #4742, #4777). Use gpu_hist instead.
    • learning_rates parameter in Python (#5155). Use the callback API instead.
    • num_roots (#5059, #5165), since the current training code always uses a single root node.
    • GPU-specific objectives (#4690), such as gpu:reg:linear. Use objectives without gpu: prefix; GPU will be used automatically if your machine has one.

    🐎 Performance improvements

    • Smarter choice of histogram construction for distributed gpu_hist (#4519)
    • Optimizations for quantization on device (#4572)
    • 🚀 Introduce caching memory allocator to avoid latency associated with GPU memory allocation (#4554, #4615)
    • ⚑️ Optimize the initialization stage of the CPU hist algorithm for sparse datasets (#4625)
    • Prevent unnecessary data copies from GPU memory to the host (#4795)
    • πŸ‘Œ Improve operation efficiency for single prediction (#5016)
    • πŸ— Group builder modified for incremental building, to speed up building large DMatrix (#5098)

    πŸ› Bug-fixes

    • πŸ—„ Eliminate FutureWarning: Series.base is deprecated (#4337)
    • Ensure pandas DataFrame column names are treated as strings in type error message (#4481)
    • πŸ“¦ [jvm-packages] Add back reg:linear for scala, as it is only deprecated and not meant to be removed yet (#4490)
    • πŸ›  Fix library loading for Cygwin users (#4499)
    • πŸ›  Fix prediction from loaded pickle (#4516)
    • Enforce exclusion between pred_interactions=True and pred_interactions=True (#4522)
    • Do not return dangling reference to local std::string (#4543)
    • Set the appropriate device before freeing device memory (#4566)
    • 0️⃣ Mark SparsePageDmatrix destructor default. (#4568)
    • Choose the appropriate tree method only when the tree method is 'auto' (#4571)
    • πŸ›  Fix (#4593)
    • πŸ“¦ [jvm-packages] Fix silly bug in feature scoring (#4604)
    • πŸ›  Fix GPU predictor when the test data matrix has different number of features than the training data matrix used to train the model (#4613)
    • πŸ›  Fix external memory for get column batches. (#4622)
    • [R] Use built-in label when xgb.DMatrix is given to (#4631)
    • πŸ›  Fix early stopping in the Python package (#4638)
    • πŸ›  Fix AUC error in distributed mode caused by imbalanced dataset (#4645, #4798)
    • πŸ“¦ [jvm-packages] Expose setMissing method in XGBoostClassificationModel / XGBoostRegressionModel (#4643)
    • βœ‚ Remove initializing stringstream reference. (#4788)
    • [R] xgb.get.handle now checks all class listed of object (#4800)
    • Do not use gpu_predictor unless data comes from GPU (#4836)
    • πŸ›  Fix data loading (#4862)
    • β†ͺ Workaround isnan across different environments. (#4883)
    • πŸ“¦ [jvm-packages] Handle Long-type parameter (#4885)
    • Don't set_params at the end of set_state (#4947). Ensure that the model does not change after pickling and unpickling multiple times.
    • C++ exceptions should not crash OpenMP loops (#4960)
    • πŸ›  Fix usegpu flag in DART. (#4984)
    • βš™ Run training with empty DMatrix (#4990, #5159)
    • πŸ›  Fix repeated split and 0 cover nodes (#5010)
    • Reset histogram hit counter between multiple data batches (#5035)
    • πŸ›  Fix feature_name crated from int64index dataframe. (#5081)
    • Don't use 0 for "fresh leaf" (#5084)
    • Throw error when user attempts to use multi-GPU training and XGBoost has not been compiled with NCCL (#5170)
    • πŸ›  Fix metric name loading (#5122)
    • Quick fix for memory leak in CPU hist algorithm (#5153)
    • πŸ›  Fix wrapping GPU ID and prevent data copying (#5160)
    • πŸ›  Fix signature of Span constructor (#5166)
    • Lazy initialization of device vector, so that XGBoost compiled with CUDA can run on a machine without any GPU (#5173)
    • Model loading should not change system locale (#5314)
    • βͺ Distributed training jobs would sometimes hang; revert Rabit to fix this regression (dmlc/rabit#132, #5237)

    API changes

    • βž• Add support for cross-validation using query ID (#4474)
    • Enable feature importance property for DART model (#4525)
    • βž• Add rmsle metric and reg:squaredlogerror objective (#4541)
    • πŸ“¦ All objective and evaluation metrics are now exposed to JVM packages (#4560)
    • dump_model() and get_dump() now support exporting in GraphViz language (#4602)
    • πŸ‘Œ Support metrics ndcg- and map- (#4635)
    • πŸ“¦ [jvm-packages] Allow chaining prediction (transform) in XGBoost4J-Spark (#4667)
    • πŸ“¦ [jvm-packages] Add option to bypass missing value check in the Spark layer (#4805). Only use this option if you know what you are doing.
    • πŸ“¦ [jvm-packages] Add public group getter (#4838)
    • πŸ—„ XGDMatrixSetGroup C API is now deprecated (#4864). Use XGDMatrixSetUIntInfo instead.
    • [R] Added new train_folds parameter to (#5114)
    • Ingest meta information from Pandas DataFrame, such as data weights (#5216)

    πŸ”¨ Maintenance: Refactor code for legibility and maintainability

    • De-duplicate GPU parameters (#4454)
    • πŸ’… Simplify INI-style config reader using C++11 STL (#4478, #4521)
    • πŸ”¨ Refactor histogram building code for gpu_hist (#4528)
    • Overload device memory allocator, to enable instrumentation for compiling memory usage statistics (#4532)
    • πŸ”¨ Refactor out row partitioning logic from gpu_hist (#4554)
    • βœ‚ Remove an unused variable (#4588)
    • Implement tree model dump with code generator, to de-duplicate code for generating dumps in 3 different formats (#4602)
    • βœ‚ Remove RowSet class which is no longer being used (#4697)
    • βœ‚ Remove some unused functions as reported by cppcheck (#4743)
    • Mimic CUDA assert output in Span check (#4762)
    • πŸ”¨ [jvm-packages] Refactor XGBoost.scala to put all params processing in one place (#4815)
    • βž• Add some comments for GPU row partitioner (#4832)
    • Span: use size_t' for index_type, addfront' and `back'. (#4935)
    • βœ‚ Remove dead code in exact algorithm (#5034, #5105)
    • Unify integer types used for row and column indices (#5034)
    • Extract feature interaction constraint from SplitEvaluator class. (#5034)
    • πŸ“„ [Breaking] De-duplicate paramters and docstrings in the constructors of Scikit-Learn models (#5130)
    • βœ‚ Remove benchmark code from GPU tests (#5141)
    • Clean up Python 2 compatibility code. (#5161)
    • Extensible binary serialization format for DMatrix::MetaInfo (#5187). This will be useful for implementing censored labels for survival analysis applications.
    • ⚠ Cleanup clang-tidy warnings. (#5247)

    🚧 Maintenance: testing, continuous integration, build system

    • πŸ‘‰ Use yaml.safe_load instead of yaml.load. (#4537)
    • Ensure GCC is at least 5.x (#4538)
    • βœ‚ Remove all mention of reg:linear from tests (#4544)
    • ⬆️ [jvm-packages] Upgrade to Scala 2.12 (#4574)
    • ⚑️ [jvm-packages] Update kryo dependency to 2.22 (#4575)
    • 🐳 [CI] Specify account ID when logging into ECR Docker registry (#4584)
    • πŸ“š Use Sphinx 2.1+ to compile documentation (#4609)
    • βœ… Make Pandas optional for running Python unit tests (#4620)
    • πŸ›  Fix spark tests on machines with many cores (#4634)
    • ⚑️ [jvm-packages] Update local dev build process (#4640)
    • βž• Add optional dependencies to (#4655)
    • πŸ“¦ [jvm-packages] Fix maven warnings (#4664)
    • βœ‚ Remove extraneous files from the R package, to comply with CRAN policy (#4699)
    • βœ‚ Remove VC-2013 support, since it is not C++11 compliant (#4701)
    • [CI] Fix broken installation of Pandas (#4704, #4722)
    • πŸ“¦ [jvm-packages] Clean up temporary files afer running tests (#4706)
    • Specify version macro in CMake. (#4730)
    • πŸ“¦ Include dmlc-tracker into XGBoost Python package (#4731)
    • πŸ–¨ [CI] Use long key ID for Ubuntu repository fingerprints. (#4783)
    • βœ‚ Remove plugin, cuda related code in automake & autoconf files (#4789)
    • βœ… Skip related tests when scikit-learn is not installed. (#4791)
    • Ignore vscode and clion files (#4866)
    • βœ… Use bundled Google Test by default (#4900)
    • ⏱ [CI] Raise timeout threshold in Jenkins (#4938)
    • Copy CMake parameter from dmlc-core. (#4948)
    • Set correct file permission. (#4964)
    • ⚑️ [CI] Update lint configuration to support latest pylint convention (#4971)
    • πŸ— [CI] Upload nightly builds to S3 (#4976, #4979)
    • βž• Add to cmake script. (#4999)
    • βœ… [CI] Fix Travis tests. (#5062)
    • [CI] Locate vcomp140.dll from System32 directory (#5078)
    • Implement training observer to dump internal states of objects (#5088). This will be useful for debugging.
    • πŸ›  Fix visual studio output library directories (#5119)
    • πŸ’… [jvm-packages] Comply with scala style convention + fix broken unit test (#5134)
    • [CI] Repair download URL for Maven 3.6.1 (#5139)
    • Don't use modernize-use-trailing-return-type in clang-tidy. (#5169)
    • Explicitly use UTF-8 codepage when using MSVC (#5197)
    • βž• Add CMake option to run Undefined Behavior Sanitizer (UBSan) (#5211)
    • βœ… Make some GPU tests deterministic (#5229)
    • πŸ— [R] Robust endian detection in CRAN xgboost build (#5232)
    • πŸ‘Œ Support FreeBSD (#5233)
    • πŸ— Make pip install xgboost*.tar.gz work by fixing (#5241)
    • πŸ›  Fix compilation error due to 64-bit integer narrowing to size_t (#5250)
    • βœ‚ Remove use of std::cout from R package, to comply with CRAN policy (#5261)
    • ⚑️ Update DMLC-Core submodule (#4674, #4688, #4726, #4924)
    • ⚑️ Update Rabit submodule (#4560, #4667, #4718, #4808, #4966, #5237)

    πŸ“š Usability Improvements, Documentation

    • βž• Add Random Forest API to Python API doc (#4500)
    • πŸ›  Fix Python demo and doc. (#4545)
    • βœ‚ Remove doc about not supporting cuda 10.1 (#4578)
    • βž• Address some sphinx warnings and errors, add doc for building doc. (#4589)
    • βž• Add instruction to run formatting checks locally (#4591)
    • πŸ›  Fix docstring for XGBModel.predict() (#4592)
    • Doc and demo for customized metric and objective (#4598, #4608)
    • βž• Add to documentation how to run tests locally (#4610)
    • Empty evaluation list in early stopping should produce meaningful error message (#4633)
    • πŸ›  Fixed year to 2019 in, helpers.h and LICENSE (#4661)
    • ⚑️ Minor updates to links and grammar (#4673)
    • βœ‚ Remove silent in doc (#4689)
    • βœ‚ Remove old Python trouble shooting doc (#4729)
    • βž• Add os.PathLike support for file paths to DMatrix and Booster Python classes (#4757)
    • ⚑️ Update XGBoost4J-Spark doc (#4804)
    • Regular formatting for evaluation metrics (#4803)
    • πŸ“š [jvm-packages] Refine documentation for handling missing values in XGBoost4J-Spark (#4805)
    • 🐎 Monitor for distributed envorinment (#4829). This is useful for identifying performance bottleneck.
    • βž• Add check for length of weights and produce a good error message (#4872)
    • πŸ›  Fix DMatrix doc (#4884)
    • Export C++ headers in CMake installation (#4897)
    • ⚑️ Update license year in to 2019 (#4940)
    • πŸ›  Fix incorrectly displayed Note in the doc (#4943)
    • πŸ“„ Follow PEP 257 Docstring Conventions (#4959)
    • βœ… Document minimum version required for Google Test (#5001)
    • βž• Add better error message for invalid feature names (#5024)
    • Some guidelines on device memory usage (#5038)
    • [doc] Some notes for external memory. (#5065)
    • ⚑️ Update document for tree_method (#5106)
    • ⚑️ Update demo for ranking. (#5154)
    • βž• Add new lines for Spark XGBoost missing values section (#5180)
    • πŸ›  Fix simple typo: utilty -> utility (#5182)
    • ⚑️ Update R doc by roxygen2 (#5201)
    • πŸ‘€ [R] Direct user to use set.seed() instead of setting seed parameter (#5125)
    • βž• Add Optuna badge to (#5208)
    • πŸ›  Fix compilation error in c-api-demo.c (#5215)

    Known issues

    • πŸ‘€ When training parameter reg_lambda is set to zero, some leaf nodes may be assigned a NaN value. (See discussion) For now, please set reg_lambda to a nonzero value.
    • πŸš€ The initial 1.0.0 release of the Python package had a small bug that rendered it incompatible with Python 3.5. See #5328. If you are using Python 3.5, install 1.0.1 instead, by running

      pip3 install xgboost==1.0.1


    Contributors : Nan Zhu (@CodingCat), Crissman Loomis (@Crissman), Cyprien Ricque (@Cyprien-Ricque), Evan Kepner (@EvanKepner), K.O. (@Hi-king), KaiJin Ji (@KerryJi), Peter Badida (@KeyWeeUsr), Kodi Arfer (@Kodiologist), Rory Mitchell (@RAMitchell), Egor Smirnov (@SmirnovEgorRu), Jacob Kim (@TheJacobKim), Vibhu Jawa (@VibhuJawa), Marcos (@astrowonk), Andy Adinets (@canonizer), Chen Qin (@chenqin), Christopher Cowden (@cowden), @cpfarrell, @david-cortes, Liangcai Li (@firestarman), @fuhaoda, Philip Hyunsu Cho (@hcho3), @here-nagini, Tong He (@hetong007), Michal Kurka (@michalkurka), Honza Sterba (@honzasterba), @iblumin, @koertkuipers, mattn (@mattn), Mingjie Tang (@merlintang), OrdoAbChao (@mglowacki100), Matthew Jones (@mt-jones), mitama (@nigimitama), Nathan Moore (@nmoorenz), Daniel Stahl (@phillyfan1138), MichaΓ«l Benesty (@pommedeterresautee), Rong Ou (@rongou), Sebastian (@sfahnens), Xu Xiao (@sperlingxx), @sriramch, Sean Owen (@srowen), Stephanie Yang (@stpyang), Yuan Tang (@terrytangyuan), Mathew Wicks (@thesuperzapper), Tim Gates (@timgates42), TinkleG (@tinkle1129), Oleksandr Pryimak (@trams), Jiaming Yuan (@trivialfis), Matvey Turkov (@turk0v), Bobby Wang (@wbo4958), yage (@yage99), @yellowdolphin

    Reviewers : Nan Zhu (@CodingCat), Crissman Loomis (@Crissman), Cyprien Ricque (@Cyprien-Ricque), Evan Kepner (@EvanKepner), John Zedlewski (@JohnZed), KOLANICH (@KOLANICH), KaiJin Ji (@KerryJi), Kodi Arfer (@Kodiologist), Rory Mitchell (@RAMitchell), Egor Smirnov (@SmirnovEgorRu), Nikita Titov (@StrikerRUS), Jacob Kim (@TheJacobKim), Vibhu Jawa (@VibhuJawa), Andrew Kane (@ankane), Arno Candel (@arnocandel), Marcos (@astrowonk), Bryan Woods (@bryan-woods), Andy Adinets (@canonizer), Chen Qin (@chenqin), Thomas Franke (@coding-komek), Peter (@codingforfun), @cpfarrell, Joshua Patterson (@datametrician), @fuhaoda, Philip Hyunsu Cho (@hcho3), Tong He (@hetong007), Honza Sterba (@honzasterba), @iblumin, @jakirkham, Vadim Khotilovich (@khotilov), Keith Kraus (@kkraus14), @koertkuipers, @melonki, Mingjie Tang (@merlintang), OrdoAbChao (@mglowacki100), Daniel Mahler (@mhlr), Matthew Rocklin (@mrocklin), Matthew Jones (@mt-jones), MichaΓ«l Benesty (@pommedeterresautee), PSEUDOTENSOR / Jonathan McKinney (@pseudotensor), Rong Ou (@rongou), Vladimir (@sh1ng), Scott Lundberg (@slundberg), Xu Xiao (@sperlingxx), @sriramch, Pasha Stetsenko (@st-pasha), Stephanie Yang (@stpyang), Yuan Tang (@terrytangyuan), Mathew Wicks (@thesuperzapper), Theodore Vasiloudis (@thvasilo), TinkleG (@tinkle1129), Oleksandr Pryimak (@trams), Jiaming Yuan (@trivialfis), Bobby Wang (@wbo4958), yage (@yage99), @yellowdolphin, Yin Lou (@yinlou)

  • v1.0.0.rc2 Changes

    February 14, 2020

    πŸ“¦ Python package

    • Linux 64-bit wheel: xgboost-1.0.0rc2-py3-none-manylinux1_x86_64.whl
    • 🏁 Windows 64-bit wheel: xgboost-1.0.0rc2-py3-none-win_amd64.whl
    • Source distribution: xgboost-1.0.0rc2.tar.gz

    πŸ“¦ R package: xgboost_1.0.0.1.tar.gz

    🐧 JVM packages (Linux 64-bit only)

    • XGBoost4J: xgboost4j_2.12-1.0.0-RC2.jar
    • XGBoost4J-Spark: xgboost4j-spark_2.12-1.0.0-RC2.jar
    • XGBoost4J-Flink: xgboost4j-flink_2.12-1.0.0-RC2.jar
  • v1.0.0.rc1 Changes

    January 31, 2020

    πŸ“¦ Python package

    • Linux 64-bit wheel: xgboost-1.0.0rc1-py2.py3-none-manylinux1_x86_64.whl
    • 🏁 Windows 64-bit wheel: xgboost-1.0.0rc1-py2.py3-none-win_amd64.whl
    • Source distribution: xgboost-1.0.0rc1.tar.gz

    πŸ“¦ R package: xgboost_1.0.0.1.tar.gz

    🐧 JVM packages (Linux 64-bit only)

    • XGBoost4J: xgboost4j_2.12-1.0.0-RC1.jar
    • XGBoost4J-Spark: xgboost4j-spark_2.12-1.0.0-RC1.jar
    • XGBoost4J-Flink: xgboost4j-flink_2.12-1.0.0-RC1.jar
  • v0.90 Changes

    May 20, 2019

    πŸ“¦ XGBoost Python package drops Python 2.x (#4379, #4381)

    πŸ“¦ Python 2.x is reaching its end-of-life at the end of this year. Many scientific Python packages are now moving to drop Python 2.x.

    XGBoost4J-Spark now requires Spark 2.4.x (#4377)

    • πŸ‘€ Spark 2.3 is reaching its end-of-life soon. See discussion at #4389.
    • Consistent handling of missing values (#4309, #4349, #4411): Many users had reported issue with inconsistent predictions between XGBoost4J-Spark and the Python XGBoost package. The issue was caused by Spark mis-handling non-zero missing values (NaN, -1, 999 etc). We now alert the user whenever Spark doesn't handle missing values correctly (#4309, #4349). See the tutorial for dealing with missing values in XGBoost4J-Spark. This fix also depends on the availability of Spark 2.4.x.

    🐎 Roadmap: better performance scaling for multi-core CPUs (#4310)

    • 🐎 Poor performance scaling of the hist algorithm for multi-core CPUs has been under investigation (#3810). #4310 optimizes quantile sketches and other pre-processing tasks. Special thanks to @SmirnovEgorRu.

    Roadmap: Harden distributed training (#4250)

    • Make distributed training in XGBoost more robust by hardening Rabit, which implements the AllReduce primitive. In particular, improve test coverage on mechanisms for fault tolerance and recovery. Special thanks to @chenqin.

    πŸ†• New feature: Multi-class metric functions for GPUs (#4368)

    • Metrics for multi-class classification have been ported to GPU: merror, mlogloss. Special thanks to @trivialfis.
    • πŸ‘ With supported metrics, XGBoost will select the correct devices based on your system and n_gpus parameter.

    πŸ†• New feature: Scikit-learn-like random forest API (#4148, #4255, #4258)

    • πŸš€ XGBoost Python package now offers XGBRFClassifier and XGBRFRegressor API to train random forests. See the tutorial. Special thanks to @canonizer

    πŸ†• New feature: use external memory in GPU predictor (#4284, #4396, #4438, #4457)

    It is now possible to make predictions on GPU when the input is read from external memory. This is useful when you want to make predictions with big dataset that does not fit into the GPU memory. Special thanks to @rongou, @canonizer, @sriramch.

    dtest = xgboost.DMatrix('test\_data.libsvm#dtest.cache') bst.set\_param('predictor', 'gpu\_predictor') bst.predict(dtest)

    Coming soon: GPU training (gpu_hist) with external memory

    πŸ†• New feature: XGBoost can now handle comments in LIBSVM files (#4430)

    πŸ†• New feature: Embed XGBoost in your C/C++ applications using CMake (#4323, #4333, #4453)

    It is now easier than ever to embed XGBoost in your C/C++ applications. In your CMakeLists.txt, add xgboost::xgboost as a linked library:

    find\_package(xgboost REQUIRED)add\_executable(api-demo c-api-demo.c)target\_link\_libraries(api-demo xgboost::xgboost)

    πŸ“š XGBoost C API documentation is available. Special thanks to @trivialfis

    🐎 Performance improvements

    • πŸ‘‰ Use feature interaction constraints to narrow split search space (#4341, #4428)
    • βž• Additional optimizations for gpu_hist (#4248, #4283)
    • ⬇️ Reduce OpenMP thread launches in gpu_hist (#4343)
    • βž• Additional optimizations for multi-node multi-GPU random forests. (#4238)
    • Allocate unique prediction buffer for each input matrix, to avoid re-sizing GPU array (#4275)
    • βœ‚ Remove various synchronisations from CUDA API calls (#4205)
    • XGBoost4J-Spark
      • Allow the user to control whether to cache partitioned training data, to potentially reduce execution time (#4268)

    πŸ› Bug-fixes

    • πŸ›  Fix node reuse in hist (#4404)
    • πŸ›  Fix GPU histogram allocation (#4347)
    • πŸ›  Fix matrix attributes not sliced (#4311)
    • Revise AUC and AUCPR metrics now work with weighted ranking task (#4216, #4436)
    • πŸ›  Fix timer invocation for InitDataOnce() in gpu_hist (#4206)
    • πŸ›  Fix R-devel errors (#4251)
    • ⚑️ Make gradient update in GPU linear updater thread-safe (#4259)
    • Prevent out-of-range access in column matrix (#4231)
    • πŸ‘» Don't store DMatrix handle in Python object until it's initialized, to improve exception safety (#4317)
    • XGBoost4J-Spark
      • Fix non-deterministic order within a zipped partition on prediction (#4388)
      • Remove race condition on tracker shutdown (#4224)
      • Allow set the parameter maxLeaves. (#4226)
      • Allow partial evaluation of dataframe before prediction (#4407)
      • Automatically set maximize_evaluation_metrics if not explicitly given (#4446)

    API changes

    • πŸ—„ Deprecate reg:linear in favor of reg:squarederror. (#4267, #4427)
    • βž• Add attribute getter and setter to the Booster object in XGBoost4J (#4336)

    πŸ”¨ Maintenance: Refactor C++ code for legibility and maintainability

    • πŸ›  Fix clang-tidy warnings. (#4149)
    • βœ‚ Remove deprecated C APIs. (#4266)
    • πŸ‘‰ Use Monitor class to time functions in hist. (#4273)
    • πŸ’… Retire DVec class in favour of c++20 style span for device memory. (#4293)
    • πŸ‘Œ Improve HostDeviceVector exception safety (#4301)

    🚧 Maintenance: testing, continuous integration, build system

    • πŸ”¨ Major refactor of CMakeLists.txt (#4323, #4333, #4453): adopt modern CMake and export XGBoost as a target
    • πŸ‘· Major improvement in Jenkins CI pipeline (#4234)
      • Migrate all Linux tests to Jenkins (#4401)
      • Builds and tests are now de-coupled, to test an artifact against multiple versions of CUDA, JDK, and other dependencies (#4401)
      • Add Windows GPU to Jenkins CI pipeline (#4463, #4469)
    • πŸ‘Œ Support CUDA 10.1 (#4223, #4232, #4265, #4468)
    • Python wheels are now built with CUDA 9.0, so that JIT is not required on Volta architecture (#4459)
    • ↔ Integrate with NVTX CUDA profiler (#4205)
    • βž• Add a test for cpu predictor using external memory (#4308)
    • πŸ”¨ Refactor tests to get rid of duplication (#4358)
    • βœ‚ Remove test dependency on craigcitro/r-travis, since it's deprecated (#4353)
    • βž• Add files from local R build to .gitignore (#4346)
    • πŸ‘‰ Make XGBoost4J compatible with Java 9+ by revising NativeLibLoader (#4351)
    • πŸ— Jenkins build for CUDA 10.0 (#4281)
    • βœ‚ Remove remaining silent and debug_verbose in Python tests (#4299)
    • 🐧 Use all cores to build XGBoost4J lib on linux (#4304)
    • ⬆️ Upgrade Jenkins Linux build environment to GCC 5.3.1, CMake 3.6.0 (#4306)
    • πŸ‘‰ Make CMakeLists.txt compatible with CMake 3.3 (#4420)
    • βž• Add OpenMP option in CMakeLists.txt (#4339)
    • ⚠ Get rid of a few trivial compiler warnings (#4312)
    • βž• Add external Docker build cache, to speed up builds on Jenkins CI (#4331, #4334, #4458)
    • πŸ›  Fix Windows tests (#4403)
    • πŸ›  Fix a broken python test (#4395)
    • πŸ‘€ Use a fixed seed to split data in XGBoost4J-Spark tests, for reproducibility (#4417)
    • βž• Add additional Python tests to test training under constraints (#4426)
    • πŸ— Enable building with shared NCCL. (#4447)

    πŸ“š Usability Improvements, Documentation

    • Document limitation of one-split-at-a-time Greedy tree learning heuristic (#4233)
    • ⚑️ Update build doc: PyPI wheel now support multi-GPU (#4219)
    • Fix docs for num_parallel_tree (#4221)
    • Fix document about colsample_by* parameter (#4340)
    • βœ… Make the train and test input with same colnames. (#4329)
    • ⚑️ Update R contribute link. (#4236)
    • πŸ›  Fix travis R tests (#4277)
    • 🌲 Log version number in crash log in XGBoost4J-Spark (#4271, #4303)
    • πŸ‘ Allow supression of Rabit output in Booster::train in XGBoost4J (#4262)
    • βž• Add tutorial on handling missing values in XGBoost4J-Spark (#4425)
    • πŸ›  Fix typos (#4345, #4393, #4432, #4435)
    • βž• Added language classifier in (#4327)
    • βž• Added Travis CI badge (#4344)
    • βž• Add BentoML to use case section (#4400)
    • βœ‚ Remove subtly sexist remark (#4418)
    • βž• Add R vignette about parsing JSON dumps (#4439)


    Contributors : Nan Zhu (@CodingCat), Adam Pocock (@Craigacp), Daniel Hen (@Daniel8hen), Jiaxiang Li (@JiaxiangBU), Rory Mitchell (@RAMitchell), Egor Smirnov (@SmirnovEgorRu), Andy Adinets (@canonizer), Jonas (@elcombato), Harry Braviner (@harrybraviner), Philip Hyunsu Cho (@hcho3), Tong He (@hetong007), James Lamb (@jameslamb), Jean-Francois Zinque (@jeffzi), Yang Yang (@jokerkeny), Mayank Suman (@mayanksuman), jess (@monkeywithacupcake), Hajime Morrita (@omo), Ravi Kalia (@project-delphi), @ras44, Rong Ou (@rongou), Shaochen Shi (@shishaochen), Xu Xiao (@sperlingxx), @sriramch, Jiaming Yuan (@trivialfis), Christopher Suchanek (@wsuchy), Bozhao (@yubozhao)

    Reviewers : Nan Zhu (@CodingCat), Adam Pocock (@Craigacp), Daniel Hen (@Daniel8hen), Jiaxiang Li (@JiaxiangBU), Laurae (@Laurae2), Rory Mitchell (@RAMitchell), Egor Smirnov (@SmirnovEgorRu), @alois-bissuel, Andy Adinets (@canonizer), Chen Qin (@chenqin), Harry Braviner (@harrybraviner), Philip Hyunsu Cho (@hcho3), Tong He (@hetong007), @jakirkham, James Lamb (@jameslamb), Julien Schueller (@jschueller), Mayank Suman (@mayanksuman), Hajime Morrita (@omo), Rong Ou (@rongou), Sara Robinson (@sararob), Shaochen Shi (@shishaochen), Xu Xiao (@sperlingxx), @sriramch, Sean Owen (@srowen), Sergei Lebedev (@superbobry), Yuan (Terry) Tang (@terrytangyuan), Theodore Vasiloudis (@thvasilo), Matthew Tovbin (@tovbinm), Jiaming Yuan (@trivialfis), Xin Yin (@xydrolase)

  • v0.82 Changes

    March 05, 2019

    πŸš€ This release is packed with many new features and bug fixes.

    🐎 Roadmap: better performance scaling for multi-core CPUs (#3957)

    • 🐎 Poor performance scaling of the hist algorithm for multi-core CPUs has been under investigation (#3810). #3957 marks an important step toward better performance scaling, by using software pre-fetching and replacing STL vectors with C-style arrays. Special thanks to @Laurae2 and @SmirnovEgorRu.
    • πŸ‘€ See #3810 for latest progress on this roadmap.

    πŸ†• New feature: Distributed Fast Histogram Algorithm (hist) (#4011, #4102, #4140, #4128)

    • It is now possible to run the hist algorithm in distributed setting. Special thanks to @CodingCat. The benefits include:
      1. Faster local computation via feature binning
      2. Support for monotonic constraints and feature interaction constraints
      3. Simpler codebase than approx, allowing for future improvement
    • πŸ”€ Depth-wise tree growing is now performed in a separate code path, so that cross-node syncronization is performed only once per level.

    πŸ†• New feature: Multi-Node, Multi-GPU training (#4095)

    • Distributed training is now able to utilize clusters equipped with NVIDIA GPUs. In particular, the rabit AllReduce layer will communicate GPU device information. Special thanks to @mt-jones, @RAMitchell, @rongou, @trivialfis, @canonizer, and @jeffdk.
    • Resource management systems will be able to assign a rank for each GPU in the cluster.
    • πŸ‘· In Dask, users will be able to construct a collection of XGBoost processes over an inhomogeneous device cluster (i.e. workers with different number and/or kinds of GPUs).

    πŸ†• New feature: Multiple validation datasets in XGBoost4J-Spark (#3904, #3910)

    • 🐎 You can now track the performance of the model during training with multiple evaluation datasets. By specifying eval_sets or call setEvalSets over a XGBoostClassifier or XGBoostRegressor, you can pass in multiple evaluation datasets typed as a Map from String to DataFrame. Special thanks to @CodingCat.
    • πŸ‘€ See the usage of multiple validation datasets here

    πŸ†• New feature: Additional metric functions for GPUs (#3952)

    • Element-wise metrics have been ported to GPU: rmse, mae, logloss, poisson-nloglik, gamma-deviance, gamma-nloglik, error, tweedie-nloglik. Special thanks to @trivialfis and @RAMitchell.
    • πŸ‘ With supported metrics, XGBoost will select the correct devices based on your system and n_gpus parameter.

    πŸ†• New feature: Column sampling at individual nodes (splits) (#3971)

    • 0️⃣ Columns (features) can now be sampled at individual tree nodes, in addition to per-tree and per-level sampling. To enable per-node sampling, set colsample_bynode parameter, which represents the fraction of columns sampled at each node. This parameter is set to 1.0 by default (i.e. no sampling per node). Special thanks to @canonizer.
    • The colsample_bynode parameter works cumulatively with other colsample_by* parameters: for example, {'colsample_bynode':0.5, 'colsample_bytree':0.5} with 100 columns will give 25 features to choose from at each split.

    🌲 Major API change: consistent logging level via verbosity (#3982, #4002, #4138)

    • ⚠ XGBoost now allows fine-grained control over logging. You can set verbosity to 0 (silent), 1 (warning), 2 (info), and 3 (debug). This is useful for controlling the amount of logging outputs. Special thanks to @trivialfis.
    • πŸ—„ Parameters silent and debug_verbose are now deprecated.
    • πŸ”§ Note: Sometimes XGBoost tries to change configurations based on heuristics, which is displayed as warning message. If there's unexpected behaviour, please try to increase value of verbosity.

    Major bug fix: external memory (#4040, #4193)

    • Clarify object ownership in multi-threaded prefetcher, to avoid memory error.
    • Correctly merge two column batches (which uses CSC layout).
    • βž• Add unit tests for external memory.
    • Special thanks to @trivialfis and @hcho3.

    πŸ›  Major bug fix: early stopping fixed in XGBoost4J and XGBoost4J-Spark (#3928, #4176)

    • πŸ“¦ Early stopping in XGBoost4J and XGBoost4J-Spark is now consistent with its counterpart in the Python package. Training stops if the current iteration is earlyStoppingSteps away from the best iteration. If there are multiple evaluation sets, only the last one is used to determinate early stop.
    • See the updated documentation here
    • Special thanks to @CodingCat, @yanboliang, and @mingyang.

    Major bug fix: infrequent features should not crash distributed training (#4045)

    • πŸ›  For infrequently occuring features, some partitions may not get any instance. This scenario used to crash distributed training due to mal-formed ranges. The problem has now been fixed.
    • In practice, one-hot-encoded categorical variables tend to produce rare features, particularly when the cardinality is high.
    • Special thanks to @CodingCat.

    🐎 Performance improvements

    • Faster, more space-efficient radix sorting in gpu_hist (#3895)
    • Subtraction trick in histogram calculation in gpu_hist (#3945)
    • More performant re-partition in XGBoost4J-Spark (#4049)

    πŸ› Bug-fixes

    • πŸ›  Fix semantics of gpu_id when running multiple XGBoost processes on a multi-GPU machine (#3851)
    • πŸ›  Fix page storage path for external memory on Windows (#3869)
    • πŸ›  Fix configuration setup so that DART utilizes GPU (#4024)
    • Eliminate NAN values from SHAP prediction (#3943)
    • Prevent empty quantile sketches in hist (#4155)
    • Enable running objectives with 0 GPU (#3878)
    • Parameters are no longer dependent on system locale (#3891, #3907)
    • πŸ‘‰ Use consistent data type in the GPU coordinate descent code (#3917)
    • βœ‚ Remove undefined behavior in the CLI config parser on the ARM platform (#3976)
    • πŸŽ‰ Initialize counters in GPU AllReduce (#3987)
    • Prevent deadlocks in GPU AllReduce (#4113)
    • Load correct values from sliced NumPy arrays (#4147, #4165)
    • πŸ›  Fix incorrect GPU device selection (#4161)
    • πŸ‘‰ Make feature binning logic in hist aware of query groups when running a ranking task (#4115). For ranking task, query groups are weighted, not individual instances.
    • 🌲 Generate correct C++ exception type for LOG(FATAL) macro (#4159)
    • πŸ“¦ Python package
      • Python package should run on system without PATH environment variable (#3845)
      • Fix coef_ and intercept_ signature to be compatible with sklearn.RFECV (#3873)
      • Use UTF-8 encoding in Python package README, to support non-English locale (#3867)
      • Add AUC-PR to list of metrics to maximize for early stopping (#3936)
      • Allow loading pickles without self.booster attribute, for backward compatibility (#3938, #3944)
      • White-list DART for feature importances (#4073)
      • Update usage of h2oai/datatable (#4123)
    • XGBoost4J-Spark
      • Address scalability issue in prediction (#4033)
      • Enforce the use of per-group weights for ranking task (#4118)
      • Fix vector size of rawPredictionCol in XGBoostClassificationModel (#3932)
      • More robust error handling in Spark tracker (#4046, #4108)
      • Fix return type of setEvalSets (#4105)
      • Return correct value of getMaxLeaves (#4114)

    API changes

    • Add experimental parameter single_precision_histogram to use single-precision histograms for the gpu_hist algorithm (#3965)
    • πŸ“¦ Python package
      • Add option to select type of feature importances in the scikit-learn inferface (#3876)
      • Add trees_to_df() method to dump decision trees as Pandas data frame (#4153)
      • Add options to control node shapes in the GraphViz plotting function (#3859)
      • Add xgb_model option to XGBClassifier, to load previously saved model (#4092)
      • Passing lists into DMatrix is now deprecated (#3970)
    • XGBoost4J
      • Support multiple feature importance features (#3801)

    πŸ”¨ Maintenance: Refactor C++ code for legibility and maintainability

    • πŸ”¨ Refactor hist algorithm code and add unit tests (#3836)
    • πŸ”¨ Minor refactoring of split evaluator in gpu_hist (#3889)
    • βœ‚ Removed unused leaf vector field in the tree model (#3989)
    • Simplify the tree representation by combining TreeModel and RegTree classes (#3995)
    • Simplify and harden tree expansion code (#4008, #4015)
    • De-duplicate parameter classes in the linear model algorithms (#4013)
    • Robust handling of ranges with C++20 span in gpu_exact and gpu_coord_descent (#4020, #4029)
    • Simplify tree training code (#3825). Also use Span class for robust handling of ranges.

    🚧 Maintenance: testing, continuous integration, build system

    • πŸ‘ Disallow std::regex since it's not supported by GCC 4.8.x (#3870)
    • βž• Add multi-GPU tests for coordinate descent algorithm for linear models (#3893, #3974)
    • πŸ’… Enforce naming style in Python lint (#3896)
    • πŸ”¨ Refactor Python tests (#3897, #3901): Use pytest exclusively, display full trace upon failure
    • βž• Address DeprecationWarning when using Python collections (#3909)
    • πŸ”Œ Use correct group for maven site plugin (#3937)
    • πŸ‘· Jenkins CI is now using on-demand EC2 instances exclusively, due to unreliability of Spot instances (#3948)
    • πŸ‘ Better GPU performance logging (#3945)
    • πŸ›  Fix GPU tests on machines with only 1 GPU (#4053)
    • ⚠ Eliminate CRAN check warnings and notes (#3988)
    • βž• Add unit tests for tree serialization (#3989)
    • βž• Add unit tests for tree fitting functions in hist (#4155)
    • βž• Add a unit test for gpu_exact algorithm (#4020)
    • Correct JVM CMake GPU flag (#4071)
    • πŸ›  Fix failing Travis CI on Mac (#4086)
    • Speed up Jenkins by not compiling CMake (#4099)
    • πŸ‘· Analyze C++ and CUDA code using clang-tidy, as part of Jenkins CI pipeline (#4034)
    • πŸ›  Fix broken R test: Install Homebrew GCC (#4142)
    • βœ… Check for empty datasets in GPU unit tests (#4151)
    • πŸ›  Fix Windows compilation (#4139)
    • πŸ‘• Comply with latest convention of cpplint (#4157)
    • πŸ›  Fix a unit test in gpu_hist (#4158)
    • βœ… Speed up data generation in Python tests (#4164)

    Usability Improvements

    • βž• Add link to InfoWorld 2019 Technology of the Year Award (#4116)
    • βœ‚ Remove outdated AWS YARN tutorial (#3885)
    • Document current limitation in number of features (#3886)
    • βœ‚ Remove unnecessary warning when gblinear is selected (#3888)
    • πŸ“œ Document limitation of CSV parser: header not supported (#3934)
    • 🌲 Log training parameters in XGBoost4J-Spark (#4091)
    • Clarify early stopping behavior in the scikit-learn interface (#3967)
    • Clarify behavior of max_depth parameter (#4078)
    • πŸ“„ Revise Python docstrings for ranking task (#4121). In particular, weights must be per-group in learning-to-rank setting.
    • Document parameter num_parallel_tree (#4022)
    • βž• Add Jenkins status badge (#4090)
    • Warn users against using internal functions of Booster object (#4066)
    • πŸ’… Reformat to comply with Python style convention (#4126)
    • Clarify a comment in objectiveTrait (#4174)
    • πŸ›  Fix typos and broken links in documentation (#3890, #3872, #3902, #3919, #3975, #4027, #4156, #4167)


    Contributors (in no particular order): Jiaming Yuan (@trivialfis), Hyunsu Cho (@hcho3), Nan Zhu (@CodingCat), Rory Mitchell (@RAMitchell), Yanbo Liang (@yanboliang), Andy Adinets (@canonizer), Tong He (@hetong007), Yuan Tang (@terrytangyuan)

    First-time Contributors (in no particular order): Jelle Zijlstra (@JelleZijlstra), Jiacheng Xu (@jiachengxu), @ajing, Kashif Rasul (@kashif), @theycallhimavi, Joey Gao (@pjgao), Prabakaran Kumaresshan (@nixphix), Huafeng Wang (@huafengw), @lyxthe, Sam Wilkinson (@scwilkinson), Tatsuhito Kato (@stabacov), Shayak Banerjee (@shayakbanerjee), Kodi Arfer (@Kodiologist), @KyleLi1985, Egor Smirnov (@SmirnovEgorRu), @tmitanitky, Pasha Stetsenko (@st-pasha), Kenichi Nagahara (@keni-chi), Abhai Kollara Dilip (@abhaikollara), Patrick Ford (@pford221), @hshujuan, Matthew Jones (@mt-jones), Thejaswi Rao (@teju85), Adam November (@anovember)

    First-time Reviewers (in no particular order): Mingyang Hu (@mingyang), Theodore Vasiloudis (@thvasilo), Jakub Troszok (@troszok), Rong Ou (@rongou), @Denisevi4, Matthew Jones (@mt-jones), Jeff Kaplan (@jeffdk)

  • v0.81 Changes

    November 04, 2018

    πŸ†• New feature: feature interaction constraints

    • πŸ‘‰ Users are now able to control which features (independent variables) are allowed to interact by specifying feature interaction constraints (#3466).
    • Tutorial is available, as well as R and Python examples.

    πŸ†• New feature: learning to rank using scikit-learn interface

    • πŸ“¦ Learning to rank task is now available for the scikit-learn interface of the Python package (#3560, #3848). It is now possible to integrate the XGBoost ranking model into the scikit-learn learning pipeline.
    • Examples of using XGBRanker class is found at demo/rank/

    πŸ†• New feature: R interface for SHAP interactions

    • πŸ“¦ SHAP (SHapley Additive exPlanations) is a unified approach to explain the output of any machine learning model. Previously, this feature was only available from the Python package; now it is available from the R package as well (#3636).

    πŸ†• New feature: GPU predictor now use multiple GPUs to predict

    • GPU predictor is now able to utilize multiple GPUs at once to accelerate prediction (#3738)

    πŸ†• New feature: Scale distributed XGBoost to large-scale clusters

    • πŸ›  Fix OS file descriptor limit assertion error on large cluster (#3835, dmlc/rabit#73) by replacing select() based AllReduce/Broadcast with poll() based implementation.
    • πŸ‘· Mitigate tracker "thundering herd" issue on large cluster. Add exponential backoff retry when workers connect to tracker.
    • With this change, we were able to scale to 1.5k executors on a 12 billion row dataset after some tweaks here and there.

    πŸ†• New feature: Additional objective functions for GPUs

    • πŸ†• New objective functions ported to GPU: hinge, multi:softmax, multi:softprob, count:poisson, reg:gamma, reg:tweedie.
    • πŸ‘ With supported objectives, XGBoost will select the correct devices based on your system and n_gpus parameter.

    Major bug fix: learning to rank with XGBoost4J-Spark

    • Previously, repartitionForData would shuffle data and lose ordering necessary for ranking task.
    • To fix this issue, data points within each RDD partition is explicitly group by their group (query session) IDs (#3654). Also handle empty RDD partition carefully (#3750).

    πŸ›  Major bug fix: early stopping fixed in XGBoost4J-Spark

    • ⚑️ Earlier implementation of early stopping had incorrect semantics and didn't let users to specify direction for optimizing (maximize / minimize)
    • A parameter maximize_evaluation_metrics is defined so as to tell whether a metric should be maximized or minimized as part of early stopping criteria (#3808). Also early stopping now has correct semantics.

    API changes

    • Column sampling by level (colsample_bylevel) is now functional for hist algorithm (#3635, #3862)
    • πŸ—„ GPU tag gpu: for regression objectives are now deprecated. XGBoost will select the correct devices automatically (#3643)
    • 0️⃣ Add disable_default_eval_metric parameter to disable default metric (#3606)
    • 🚚 Experimental AVX support for gradient computation is removed (#3752)
    • XGBoost4J-Spark
      • Add rank:ndcg and rank:map to supported objectives (#3697)
    • πŸ“¦ Python package
      • Add callbacks argument to fit() function of sciki-learn API (#3682)
      • Add XGBRanker to scikit-learn interface (#3560, #3848)
      • Add validate_features argument to predict() function of scikit-learn API (#3653)
      • Allow scikit-learn grid search over parameters specified as keyword arguments (#3791)
      • Add coef_ and intercept_ as properties of scikit-learn wrapper (#3855). Some scikit-learn functions expect these properties.

    🐎 Performance improvements

    • βž• Address very high GPU memory usage for large data (#3635)
    • πŸ›  Fix performance regression within EvaluateSplits() of gpu_hist algorithm. (#3680)

    πŸ› Bug-fixes

    • πŸ›  Fix a problem in GPU quantile sketch with tiny instance weights. (#3628)
    • πŸ›  Fix copy constructor for HostDeviceVectorImpl to prevent dangling pointers (#3657)
    • πŸ›  Fix a bug in partitioned file loading (#3673)
    • πŸ›  Fixed an uninitialized pointer in gpu_hist (#3703)
    • Reshared data among GPUs when number of GPUs is changed (#3721)
    • Add back max_delta_step to split evaluation (#3668)
    • Do not round up integer thresholds for integer features in JSON dump (#3717)
    • πŸ‘‰ Use dmlc::TemporaryDirectory to handle temporaries in cross-platform way (#3783)
    • Fix accuracy problem with gpu_hist when min_child_weight and lambda are set to 0 (#3793)
    • πŸ‘‰ Make sure that tree_method parameter is recognized and not silently ignored (#3849)
    • XGBoost4J-Spark
      • Make sure thresholds are considered when executing predict() method (#3577)
      • Avoid losing precision when computing probabilities by converting to Double early (#3576)
      • getTreeLimit() should return Int (#3602)
      • Fix checkpoint serialization on HDFS (#3614)
      • Throw ControlThrowable instead of InterruptedException so that it is properly re-thrown (#3632)
      • Remove extraneous output to stdout (#3665)
      • Allow specification of task type for custom objectives and evaluations (#3646)
      • Fix distributed updater check (#3739)
      • Fix issue when spark job execution thread cannot return before we execute first() (#3758)
    • πŸ“¦ Python package
      • Fix accessing DMatrix.handle before it is set (#3599)
      • XGBClassifier.predict() should return margin scores when output_margin is set to true (#3651)
      • Early stopping callback should maximize metric of form [email protected] (#3685)
      • Preserve feature names when slicing DMatrix (#3766)
    • πŸ“¦ R package
      • Replace nround with nrounds to match actual parameter (#3592)
      • Amend xgb.createFolds to handle classes of a single element (#3630)
      • Fix buggy random generator and make colsample_bytree functional (#3781)

    🚧 Maintenance: testing, continuous integration, build system

    • βž• Add sanitizers tests to Travis CI (#3557)
    • βž• Add NumPy, Matplotlib, Graphviz as requirements for doc build (#3669)
    • Comply with CRAN submission policy (#3660, #3728)
    • βœ‚ Remove copy-paste error in JVM test suite (#3692)
    • ⚑️ Disable flaky tests in R-package/tests/testthat/test_update.R (#3723)
    • πŸš€ Make Python tests compatible with scikit-learn 0.20 release (#3731)
    • πŸ— Separate out restricted and unrestricted tasks, so that pull requests don't build downloadable artifacts (#3736)
    • βž• Add multi-GPU unit test environment (#3741)
    • πŸ‘ Allow plug-ins to be built by CMake (#3752)
    • βœ… Test wheel compatibility on CPU containers for pull requests (#3762)
    • πŸ›  Fix broken doc build due to Matplotlib 3.0 release (#3764)
    • Produce for XGBoost-R on Mac OSX, so that make install works (#3767)
    • πŸ‘· Retry Jenkins CI tests up to 3 times to improve reliability (#3769, #3769, #3775, #3776, #3777)
    • βž• Add basic unit tests for gpu_hist algorithm (#3785)
    • πŸ›  Fix Python environment for distributed unit tests (#3806)
    • βœ… Test wheels on CUDA 10.0 container for compatibility (#3838)
    • πŸ›  Fix JVM doc build (#3853)

    πŸ”¨ Maintenance: Refactor C++ code for legibility and maintainability

    • πŸ”€ Merge generic device helper functions into GPUSet class (#3626)
    • Re-factor column sampling logic into ColumnSampler class (#3635, #3637)
    • πŸ“œ Replace std::vector with HostDeviceVector in MetaInfo and SparsePage (#3446)
    • Simplify DMatrix class (#3395)
    • De-duplicate CPU/GPU code using Transform class (#3643, #3751)
    • βœ‚ Remove obsoleted QuantileHistMaker class (#3761)
    • βœ‚ Remove obsoleted NoConstraint class (#3792)

    Other Features

    • C++20-compliant Span class for safe pointer indexing (#3548, #3588)
    • βž• Add helper functions to manipulate multiple GPU devices (#3693)
    • XGBoost4J-Spark
      • Allow specifying host ip from the file (#3833). This comes in handy when hosts files doesn't correctly define localhost.

    Usability Improvements

    • βž• Add reference to GitHub repository in pom.xml of JVM packages (#3589)
    • βž• Add R demo of multi-class classification (#3695)
    • Document JSON dump functionality (#3600, #3603)
    • Document CUDA requirement and lack of external memory for GPU algorithms (#3624)
    • Document LambdaMART objectives, both pairwise and listwise (#3672)
    • Document aucpr evaluation metric (#3687)
    • Document gblinear parameters: feature_selector and top_k (#3780)
    • βž• Add instructions for using MinGW-built XGBoost with Python. (#3774)
    • βœ‚ Removed nonexistent parameter use_buffer from documentation (#3610)
    • ⚑️ Update Python API doc to include all classes and members (#3619, #3682)
    • πŸ›  Fix typos and broken links in documentation (#3618, #3640, #3676, #3713, #3759, #3784, #3843, #3852)
    • Binary classification demo should produce LIBSVM with 0-based indexing (#3652)
    • πŸ–¨ Process data once for Python and CLI examples of learning to rank (#3666)
    • Include full text of Apache 2.0 license in the repository (#3698)
    • πŸ’Ύ Save predictor parameters in model file (#3856)
    • πŸ“¦ JVM packages
      • Let users specify feature names when calling getModelDump and getFeatureScore (#3733)
      • Warn the user about the lack of over-the-wire encryption (#3667)
      • Fix errors in examples (#3719)
      • Document choice of trackers (#3831)
      • Document that vanilla Apache Spark is required (#3854)
    • πŸ“¦ Python package
      • Document that custom objective can't contain colon (:) (#3601)
      • Show a better error message for failed library loading (#3690)
      • Document that feature importance is unavailable for non-tree learners (#3765)
      • Document behavior of get_fscore() for zero-importance features (#3763)
      • Recommend pickling as the way to save XGBClassifier / XGBRegressor / XGBRanker (#3829)
    • πŸ“¦ R package
      • Enlarge variable importance plot to make it more visible (#3820)


    • ⬆️ External memory page files have changed, breaking backwards compatibility for temporary storage used during external memory training. This only affects external memory users upgrading their xgboost version - we recommend clearing all *.page files before resuming training. Model serialization is unaffected.

    Known issues

    • Quantile sketcher fails to produce any quantile for some edge cases (#2943)
    • The hist algorithm leaks memory when used with learning rate decay callback (#3579)
    • Using custom evaluation funciton together with early stopping causes assertion failure in XGBoost4J-Spark (#3595)
    • Early stopping doesn't work with gblinear learner (#3789)
    • Label and weight vectors are not reshared upon the change in number of GPUs (#3794). To get around this issue, delete the DMatrix object and re-load.
    • The DMatrix Python objects are initialized with incorrect values when given array slices (#3841)
    • πŸ‘ The gpu_id parameter is broken and not yet properly supported (#3850)


    Contributors (in no particular order): Hyunsu Cho (@hcho3), Jiaming Yuan (@trivialfis), Nan Zhu (@CodingCat), Rory Mitchell (@RAMitchell), Andy Adinets (@canonizer), Vadim Khotilovich (@khotilov), Sergei Lebedev (@superbobry)

    First-time Contributors (in no particular order): Matthew Tovbin (@tovbinm), Jakob Richter (@jakob-r), Grace Lam (@grace-lam), Grant W Schneider (@grantschneider), Andrew Thia (@BlueTea88), Sergei Chipiga (@schipiga), Joseph Bradley (@jkbradley), Chen Qin (@chenqin), Jerry Lin (@linjer), Dmitriy Rybalko (@rdtft), Michael Mui (@mmui), Takahiro Kojima (@515hikaru), Bruce Zhao (@BruceZhaoR), Wei Tian (@weitian), Saumya Bhatnagar (@Sam1301), Juzer Shakir (@JuzerShakir), Zhao Hang (@cleghom), Jonathan Friedman (@jontonsoup), Bruno Tremblay (@meztez), Boris Filippov (@frenzykryger), @Shiki-H, @mrgutkun, @gorogm, @htgeis, @jakehoare, @zengxy, @KOLANICH

    First-time Reviewers (in no particular order): Nikita Titov (@StrikerRUS), Xiangrui Meng (@mengxr), Nirmal Borah (@Nirmal-Neel)

  • v0.80 Changes

    August 13, 2018
    • ⬆️ JVM packages received a major upgrade : To consolidate the APIs and improve the user experience, we refactored the design of XGBoost4J-Spark in a significant manner. (#3387)
      • Consolidated APIs: It is now much easier to integrate XGBoost models into a Spark ML pipeline. Users can control behaviors like output leaf prediction results by setting corresponding column names. Training is now more consistent with other Estimators in Spark MLLIB: there is now one single method fit() to train decision trees.
      • Better user experience: we refactored the parameters relevant modules in XGBoost4J-Spark to provide both camel-case (Spark ML style) and underscore (XGBoost style) parameters
      • A brand-new tutorial is available for XGBoost4J-Spark.
      • Latest API documentation is now hosted at
    • πŸ“š XGBoost documentation now keeps track of multiple versions:
    • πŸ‘Œ Support for per-group weights in ranking objective (#3379)
    • πŸ›  Fix inaccurate decimal parsing (#3546)
    • πŸ†• New functionality
      • Query ID column support in LIBSVM data files (#2749). This is convenient for performing ranking task in distributed setting.
      • Hinge loss for binary classification (binary:hinge) (#3477)
      • Ability to specify delimiter and instance weight column for CSV files (#3546)
      • Ability to use 1-based indexing instead of 0-based (#3546)
    • πŸ‘ GPU support
      • Quantile sketch, binning, and index compression are now performed on GPU, eliminating PCIe transfer for 'gpu_hist' algorithm (#3319, #3393)
      • Upgrade to NCCL2 for multi-GPU training (#3404).
      • Use shared memory atomics for faster training (#3384).
      • Dynamically allocate GPU memory, to prevent large allocations for deep trees (#3519)
      • Fix memory copy bug for large files (#3472)
    • πŸ“¦ Python package
      • Importing data from Python datatable (#3272)
      • Pre-built binary wheels available for 64-bit Linux and Windows (#3424, #3443)
      • Add new importance measures 'total_gain', 'total_cover' (#3498)
      • Sklearn API now supports saving and loading models (#3192)
      • Arbitrary cross validation fold indices (#3353)
      • predict() function in Sklearn API uses best_ntree_limit if available, to make early stopping easier to use (#3445)
      • Informational messages are now directed to Python's print() rather than standard output (#3438). This way, messages appear inside Jupyter notebooks.
    • πŸ“¦ R package
      • Oracle Solaris support, per CRAN policy (#3372)
    • πŸ“¦ JVM packages
      • Single-instance prediction (#3464)
      • Pre-built JARs are now available from Maven Central (#3401)
      • Add NULL pointer check (#3021)
      • Consider spark.task.cpus when controlling parallelism (#3530)
      • Handle missing values in prediction (#3529)
      • Eliminate outputs of System.out (#3572)
    • πŸ”¨ Refactored C++ DMatrix class for simplicity and de-duplication (#3301)
    • πŸ”¨ Refactored C++ histogram facilities (#3564)
    • πŸ”¨ Refactored constraints / regularization mechanism for split finding (#3335, #3429). Users may specify an elastic net (L2 + L1 regularization) on leaf weights as well as monotonic constraints on test nodes. The refactor will be useful for a future addition of feature interaction constraints.
    • Statically link libstdc++ for MinGW32 (#3430)
    • πŸ‘€ Enable loading from group, base_margin and weight (see here) for Python, R, and JVM packages (#3431)
    • Fix model saving for count:possion so that max_delta_step doesn't get truncated (#3515)
    • πŸ›  Fix loading of sparse CSC matrix (#3553)
    • πŸ›  Fix incorrect handling of base_score parameter for Tweedie regression (#3295)
  • v0.72 Changes

    June 01, 2018
    • πŸš€ Starting with this release, we plan to make a new release every two months. See #3252 for more details.
    • πŸ›  Fix a pathological behavior (near-zero second-order gradients) in multiclass objective (#3304)
    • Tree dumps now use high precision in storing floating-point values (#3298)
    • πŸ›  Submodules rabit and dmlc-core have been brought up to date, bringing bug fixes (#3330, #3221).
    • πŸ‘ GPU support
      • Continuous integration tests for GPU code (#3294, #3309)
      • GPU accelerated coordinate descent algorithm (#3178)
      • Abstract 1D vector class now works with multiple GPUs (#3287)
      • Generate PTX code for most recent architecture (#3316)
      • Fix a memory bug on NVIDIA K80 cards (#3293)
      • Address performance instability for single-GPU, multi-core machines (#3324)
    • πŸ“¦ Python package
      • FreeBSD support (#3247)
      • Validation of feature names in Booster.predict() is now optional (#3323)
    • ⚑️ Updated Sklearn API
      • Validation sets now support instance weights (#2354)
      • XGBClassifier.predict_proba() should not support output_margin option. (#3343) See BREAKING CHANGES below.
    • πŸ“¦ R package:
      • Better handling of NULL in print.xgb.Booster() (#3338)
      • Comply with CRAN policy by removing compiler warning suppression (#3329)
      • Updated CRAN submission
    • πŸ“¦ JVM packages
      • JVM packages will now use the same versioning scheme as other packages (#3253)
      • Update Spark to 2.3 (#3254)
      • Add scripts to cross-build and deploy artifacts (#3276, #3307)
      • Fix a compilation error for Scala 2.10 (#3332)
      • XGBClassifier.predict_proba() no longer accepts paramter output_margin. The paramater makes no sense for predict_proba() because the method is to predict class probabilities, not raw margin scores.