v2.0.0

Breaking changes:

• ordering of the parameters and their defaults now matching with the official LightGBM documentation

• some parameter names have been changed to remove non-aliases, e.g. is_sparse is now is_enable_sparse

• added verbosity controllable on model level (previously a missing parameter) and changed its default from -1 to 1

Added:

• added all previously missing parameters (except CLI only params), e.g. min_data_in_bin, refit_decay_rate
• implemented missing C-API functions: LGBM_BoosterPredictForFile, LGBM_DatasetCreateFromFile, LGBM_BoosterRefit, LGBM_BoosterGetLinear
• enabled refitting supporting refit_decay_rate param
• implemented missing values support (in core LightGBM)

patch release v1.0.1

• added MLJ compliant docstrings

v1.0.0

Release v1.0.0

• updated all currently implemented parameters with their non-alias names and defaults as per lightgbm C-API documentation v3.3.5 (breaking change for any code referencing parameter names prior this change, e.g. is_training_metric which is now is_provide_training_metric)

v0.8.0

• added new estimator LGBMRanking with defaulted objective "lambdarank"
• fixed empty cv object when no metric is provided
• added tests to LGBM_DatasetSetField

patch release v0.7.2

• fixed linear_tree param order in LGBMRegression

patch release v0.7.1

• fixed mutation of categorical features indices on multiple stringifyparams calls
• added explicit objective=none to LGBM_BoosterUpdateOneIterCustom

v0.7.0

Release v0.7.0:

• added LightGBM_jll instead of pre-compiled binaries
• added support for macos-14
• added conditional download of pre-compiled binaries for Julia 1.8.4 and 1.8.5 for WIndows
• changed:
  • julia compat to >=1.6
  • MLJBase compat >= 1.0
• removed support for julia < 1.6

patch release v0.6.2

• added linear_tree param
• extended CI matrix including MLJ tests to julia 1.10
• fixed docs build

patch release v0.6.1

• Added the following to MLJModelInterface:
  • gpu_use_dp
  • gpu_platform_id
  • gpu_device_id
  • num_gpu
• Added num_threads for predict
• Changed LightGBM examples to v3.3.5

v0.6.0

What's Changed

• Adding support for constructing Dataset from SparseMatrix by @yaxxie in #113
• adding support for LGBM_BoosterUpdateOneIterCustom by @yaxxie in #114
• small fix to custom objective test by @yaxxie in #115
• Update to latest bundled binaries by @yaxxie in #119
• Intermittent test fail in Custom objective tests by @yaxxie in #120
• Occursin test for later boosters by @yaxxie in #121
• Fix failing tests by @FatemehTahavori in #126
• Don't try find_library tests if the system doesn't satisfy the assumptions by @yaxxie in #128

New Contributors

• @yaxxie made their first contribution in #113
• @FatemehTahavori made their first contribution in #126

Full Changelog: v0.5.2...v0.6.0