Releases · microsoft/LightGBM · GitHub

Changes 💡 New Features [CUDA] fix setting of CUDA architectures and enable support for NVIDIA Blackwell @StrikerRUS (#6812) [python-package] support Python 3.13 @jameslamb (#6668) [GPU] Add support for linear tree with device=gpu @dragonbra (#6567) [c++] Add Bagging by Query for Lambdarank @shiyu1994 (#6623) 🔨 Breaking [R-package] require lgb.Dataset, remove support for passing 'colnames' and 'categorical_feature' for lgb.train() and lgb.cv() @jameslamb (#6714) [python-package] support sub-classing scikit-learn estimators @jameslamb (#6783) [python-package] do not copy column-major numpy arrays when creating Dataset from list of arrays @jmoralez (#6773) [python-package] remove support for passing 'feature_name' and 'categorical_feature' through train() and cv() @jameslamb (#6706) [python-package] require scikit-learn>=0.24.2, make scikit-learn estimators compatible with scikit-learn>=1.6.0dev @vnherdeiro (#6651) [ci] Require CMake 3.28 and replace FetchContent_Populate with FetchContent_MakeAvailable @StrikerRUS (#6550) 🚀 Efficiency Improvement [python-package] do not copy column-major numpy arrays when predicting @jmoralez (#6751) [python-package] do not copy column-major numpy arrays when creating Dataset @jmoralez (#6721) 🐛 Bug Fixes [python-package] Separately check whether pyarrow and cffi are installed @mlondschien (#6785) [c++] Fixed Predictor lifecycle and trees initialization in Contrib mode @AndreyOrb (#6778) [python-package] Infer feature names from pyarrow.Table @mlondschien (#6781) [python-package] Fix inconsistency in predict() output shape for 1-tree models @RektPunk (#6753) [fix] resolve potential attack in linker connection building @shiyu1994 (#6752) [R-package] Avoid bashisms (non-POSIX code) in R-package/configure @smoser (#6746) [c++] fix parallel_tree_learner_split_info @moming39 (#6738) [c++] Fix dump_model() information for root node @neNasko1 (#6569) [cmake] [R-package] include R-for-macOS vendored libs dir in OpenMP search path (fixes #6628