LightGBM RankingΒΆ

The documentation is generated based on the sources available at dotnet/machinelearning and released under MIT License.

Type: rankertrainer Aliases: LightGBMRanking, LightGBMRank Namespace: Microsoft.ML.Runtime.LightGBM Assembly: Microsoft.ML.LightGBM.dll Microsoft Documentation: LightGBM Ranking


LightGBM Ranking


Name Short name Default Description
batchSize   1048576 Number of entries in a batch when loading data.
booster   LightGbmArguments.TreeBooster.Arguments Which booster to use, can be gbtree, gblinear or dart. gbtree and dart use tree based model while gblinear uses linear function.
catL2   10 L2 Regularization for categorical split.
catSmooth   10 Lapalace smooth term in categorical feature spilt. Avoid the bias of small categories.
customGains gains 0, 3, 7, 15, 31, 63, 127, 255, 511, 1023, 2047, 4095 Comma seperated list of gains associated to each relevance label.
earlyStoppingRound es 0 Rounds of early stopping, 0 will disable it.
evalMetric em DefaultMetric Evaluation metrics.
learningRate lr   Shrinkage rate for trees, used to prevent over-fitting. Range: (0,1].
maxBin mb 255 Max number of bucket bin for features.
maxCatThreshold maxcat 32 Max number of categorical thresholds.
minDataPerGroup mdpg 100 Min number of instances per categorical group.
minDataPerLeaf mil   Minimum number of instances needed in a child.
nThread nt   Number of parallel threads used to run LightGBM.
numBoostRound iter 100 Number of iterations.
numLeaves nl   Maximum leaves for trees.
parallelTrainer parag Microsoft. ML. Runtime. LightGBM. SingleTrainerFactory Parallel LightGBM Learning Algorithm
sigmoid   0.5 Parameter for the sigmoid function. Used only in LightGbmBinaryTrainer, LightGbmMulticlassTrainer and in LightGbmRankingTrainer.
silent   True Printing running messages.
useCat cat   Enable categorical split or not.
useMissing   False Enable missing value auto infer or not.
useSoftmax     Use softmax loss for the multi classification.
verboseEval v False Verbose