A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.
✨ v4.2.0 of the R package is now available on CRAN (link), the first major release of the R package in 2+ years.
✨ The Python package now accepts Apache Arrow Tables and Arrays (thanks @borchero!)
🔧 A critical bug in quantized training support is fixed
random_state @david-cortes (#6174)parallel region @jameslamb (#6135)LGBMClassifier.fit @david-cortes (#6002)cat() instead of print() for metrics and callbacks @david-cortes (#6171)mamba instead of conda in macOS and Linux CI jobs @borchero (#6140)lgb.Dataset instance to lightgbm() @david-cortes (#6005)Note
This release of the R package was not submitted to CRAN, due to the issues documented in https://github.com/microsoft/LightGBM/issues/5987.
This release contains all previously-unreleased changes since v3.3.1 over 1.5 years ago (link).
Summary of improvements:
scikit-build-core (link) as its build backendmanylinux_2_28 Linux wheels now support GPU (OpenCL-based, not CUDA) build automatically... just pip install lightgbm then pass {"device": "gpu"} in params (thanks @jgiannuzzi!)py.typed so any code using LightGBM can benefitpandas nullable typeslgb.early_stopping(..., min_delta=n)) for how much eval metrics must improve to be considered "improved" for early stoppingscikit-learn is no longer a required dependencyray, Dask) (thanks @Yard1!)dgCMatrix and dsparseMatrix (thanks @david-cortes!)saveRDS() and readRDS() for Booster, print() and summary() methods for Dataset (thanks @david-cortes!)Summary of breaking changes:
from lightgbm import *
setup.py, pip install --install-option supportpip install --install-option (to work with newer pip, see https://github.com/pypa/pip/issues/11358)
MSBUild.exe ... that now requires compiling lib_lightgbm.dll separately and then building a wheel that bundles it...
lgb.unloader()
predict(newdata, type = ...) in predict(), for consistency with base R and most other machine learning projectsplot_tree and create_tree_digraph (fixes #4784) @jmoralez (#5119)predict @david-cortes (#4977)reset_parameter callback pickleable @StrikerRUS (#5109)record_evaluation callback pickleable @StrikerRUS (#5107)log_evaluation callback pickleable @StrikerRUS (#5101)weight -> weights @david-cortes (#4975)early_stopping callback pickleable @Yard1 (#5012)lightgbm() @david-cortes (#4976)init() and set_params() methods @StrikerRUS (#4822)print() and summary() methods for Booster @david-cortes (#4686)n_estimators_ and n_iter_ post-fit attributes @StrikerRUS (#4753)setup.py) (fixes #5061) @jameslamb (#5759)type argument to control prediction types @david-cortes (#5133)n_jobs and change default to number of cores @david-cortes (#5105)reshape argument in predict @david-cortes (#4971)lightgbm() and change default to number of cores @david-cortes (#4972)data -> newdata in predict @david-cortes (#4973)windows-2019 image instead of vs2017-win2016 @StrikerRUS (#5059)early_stopping_rounds argument of train() and cv() functions @StrikerRUS (#4908)evals_result argument of train() function @StrikerRUS (#4882)None for evals_result_ @StrikerRUS (#4884)verbose_eval argument of train() and cv() functions @StrikerRUS (#4878)verbose argument of model_from_string() method of Booster class @StrikerRUS (#4877)early_stopping_rounds argument of fit() method @StrikerRUS (#4846)slice() @jameslamb (#4872)lgb.Dataset() @jameslamb (#4874)dim.lgb.Dataset() @jameslamb (#4873)create_valid() @jameslamb (#4865)best_iteration for sklearn and standard APIs @StrikerRUS (#4845)verbose argument from fit() method @StrikerRUS (#4832)learning_rates argument of train() function @StrikerRUS (#4831)ylabel argument of plot_metric() function @StrikerRUS (#4818)print_evaluation() function @StrikerRUS (#4819)silent argument @StrikerRUS (#4800)delete[] where appropriate instead of delete @david-cortes (#4984)print.lgb.Booster() @StrikerRUS (#4941)--no-build-vignettes option for build-cran-package.sh @jameslamb (#4848)array-like wording @StrikerRUS (#4816)time-costs Python-package build option @StrikerRUS (#5554)conda-forge channel preference over default one and describe possible workaround for OpenMP conflicts in FAQ @StrikerRUS (#4994)pred_early_stop can be used only in normal and raw scores prediction @StrikerRUS (#4823)nthreads parameter @StrikerRUS (#4756)Dataset.feature_penalty @jameslamb (#5656)ubuntu-latest at CI (fixes #5186) @StrikerRUS (#5288)pandas.Series.iteritems with pandas.Series.items @jmoralez (#5506)#ifndef guards @StrikerRUS (#5466)11.7.0 to 11.7.1 at CI @StrikerRUS (#5442)11.6.2 to 11.7.0 at CI @StrikerRUS (#5287)basic.is_numeric() with _ @ak04p (#5421)lintr code @StrikerRUS (#5327)predict() with _choose_param_value() @StrikerRUS (#5308)nolint mark @StrikerRUS (#5300)basic.py @StrikerRUS (#5188)fobj with custom objective in test comments and make tests stricter @StrikerRUS (#5173)11.5.1 to 11.6.2 at CI @StrikerRUS (#5149)master fails with graphviz-related error @StrikerRUS (#5068)train() function defined in engine.py from test_basic.py to test_engine.py @StrikerRUS (#5034)bdist CI job in old Ubuntu container @StrikerRUS (#5022)latest tag again for Ubuntu 14 CI Docker @StrikerRUS (#5001)11.5.0 to 11.5.1 at CI @StrikerRUS (#4967)-1 for exit code in case of failure @StrikerRUS (#4939)# symbol for comments @StrikerRUS (#4940)testthat::expect_is() @jameslamb (#4916)UnboundLocalError in early stopping callback for Python 2 @StrikerRUS (#4855)_compare_params_for_warning() and make it reusable @StrikerRUS (#4824)CMAKE_CUDA_FLAGS variable in CMakeLists.txt @StrikerRUS (#4799)linelength errors @StrikerRUS (#4796)endif() and endfunction() without expression inside in CMakeLists.txt @StrikerRUS (#4798)filename that is no longer true with new version of graphviz @StrikerRUS (#4778)freetype version @StrikerRUS (#4776)cv() function @StrikerRUS (#4766)datatable to the mocked modules during docs building process and sort them alphabetically @StrikerRUS (#4750)This is a special release, put up to prevent the R package from being archived on CRAN.
See #5661 and #5662 for context.
This release only contains the changes, relative to v3.3.4, necessary to prevent removal of the R package from CRAN.
None
None
None
None
None
This is a special release, put up to prevent the R package from being archived on CRAN.
See #5618 and #5619 for context.
This release only contains the changes, relative to v3.3.3, necessary to prevent removal of the R package from CRAN.
None
None
None
None
This is a special release, put up to prevent the R package from being archived on CRAN.
See https://github.com/microsoft/LightGBM/issues/5502 and https://github.com/microsoft/LightGBM/pull/5525 for context.
This release only contains the changes, relevant to v3.3.2, necessary to prevent removal of the R package from CRAN.
None
None
None
std::function usage @david-cortes (#4673)1.10.0 to 1.11.0 @StrikerRUS (#4683)setup.py @StrikerRUS (#4663)licenseUrl field with license one in .nuspec file @StrikerRUS (#4669)__sklearn_is_fitted__() in all estimator fitness checks @StrikerRUS (#4654)init_score in multiclass classification task @jmoralez (#4150)ylabel argument of plot_metric() function @StrikerRUS (#4624)print_evaluation() into log_evaluation() @StrikerRUS (#4604)train() and cv() functions and sklearn wrapper @StrikerRUS (#4574)silent and standalone verbose args. Prefer global verbose param @StrikerRUS (#4577)importance_type param in plotting @StrikerRUS (#4570)eval_at aliases in keyword arguments @StrikerRUS (#4599)callable in docstrings @StrikerRUS (#4575)str in docstrings @StrikerRUS (#4565)str and add commas in list of ... types @StrikerRUS (#4557)eval_result argument in record_evaluation() @StrikerRUS (#4559)daal4py in README @StrikerRUS (#4532)keep_training_booster param description @StrikerRUS (#4364)__sklearn_is_fitted__() method to be better compatible with scikit-learn API @StrikerRUS (#4636)CMakeLists.txt for MPI @StrikerRUS (#4644)setup.py @StrikerRUS (#4620)dask.py @StrikerRUS (#4615)init_score array when set by predictor @StrikerRUS (#4510)CreateSampleIndices() in c_api.cpp @cyfdecyf (#4478)absolute() on paths first @StrikerRUS (#4444).gitignore @StrikerRUS (#4429)prints in simple_example.py and sklearn_example.py @StrikerRUS (#4396)