mlrl.boosting.config.parameters module

Author: Michael Rapp (michael.rapp.ml@gmail.com)

Provides utility function for configuring boosting algorithms.

class mlrl.boosting.config.parameters.BinaryPredictorParameter

Bases: NominalParameter

A parameter that allows to configure the strategy to be used for predicting binary labels.

BINARY_PREDICTOR_EXAMPLE_WISE = 'example-wise'

BINARY_PREDICTOR_GFM = 'gfm'

BINARY_PREDICTOR_OUTPUT_WISE = 'output-wise'

class mlrl.boosting.config.parameters.ClassificationLossParameter

Bases: RegressionLossParameter

A parameter that allows to configure the loss function to be minimized during training in classification problems.

LOSS_LOGISTIC_DECOMPOSABLE = 'logistic-decomposable'

LOSS_LOGISTIC_NON_DECOMPOSABLE = 'logistic-non-decomposable'

LOSS_SQUARED_HINGE_DECOMPOSABLE = 'squared-hinge-decomposable'

LOSS_SQUARED_HINGE_NON_DECOMPOSABLE = 'squared-hinge-non-decomposable'

class mlrl.boosting.config.parameters.DefaultRuleParameter

Bases: NominalParameter

A parameter that allows to configure whether a default rule should be induced or not.

class mlrl.boosting.config.parameters.ExtendedFeatureBinningParameter

Bases: FeatureBinningParameter

Extends the FeatureBinningParameter by a value for automatic configuration.

class mlrl.boosting.config.parameters.ExtendedParallelRuleRefinementParameter

Bases: ParallelRuleRefinementParameter

Extends the ParallelRuleRefinementParameter by a value for automatic configuration.

class mlrl.boosting.config.parameters.ExtendedParallelStatisticUpdateParameter

Bases: ParallelStatisticUpdateParameter

Extends the ParallelStatisticUpdateParameter by a value for automatic configuration.

class mlrl.boosting.config.parameters.ExtendedPartitionSamplingParameter

Bases: PartitionSamplingParameter

Extends the PartitionSamplingParameter by a value for automatic configuration.

class mlrl.boosting.config.parameters.HeadTypeParameter

Bases: NominalParameter

A parameter that allows to configure the type of the rule heads that should be used.

HEAD_TYPE_COMPLETE = 'complete'

HEAD_TYPE_PARTIAL_DYNAMIC = 'partial-dynamic'

HEAD_TYPE_PARTIAL_FIXED = 'partial-fixed'

HEAD_TYPE_SINGLE = 'single'

OPTION_EXPONENT = 'exponent'

OPTION_MAX_OUTPUTS = 'max_outputs'

OPTION_MIN_OUTPUTS = 'min_outputs'

OPTION_OUTPUT_RATIO = 'output_ratio'

OPTION_THRESHOLD = 'threshold'

class mlrl.boosting.config.parameters.JointProbabilityCalibrationParameter

Bases: NominalParameter

A parameter that allows to configure the method to be used for the calibration of joint probabilities.

class mlrl.boosting.config.parameters.L1RegularizationParameter

Bases: FloatParameter

A parameter that allows to configure the weight of the L1 regularization.

class mlrl.boosting.config.parameters.L2RegularizationParameter

Bases: FloatParameter

A parameter that allows to configure the weight of the L2 regularization.

class mlrl.boosting.config.parameters.LabelBinningParameter

Bases: NominalParameter

A parameter that allows to configure the strategy to be used for gradient-based label binning (GBLB).

class mlrl.boosting.config.parameters.MarginalProbabilityCalibrationParameter

Bases: NominalParameter

A parameter that allows to configure the method to be used for the calibration of marginal probabilities.

class mlrl.boosting.config.parameters.ProbabilityPredictorParameter

Bases: NominalParameter

A parameter that allows to configure the strategy to be used for predicting probabilities.

PROBABILITY_PREDICTOR_MARGINALIZED = 'marginalized'

PROBABILITY_PREDICTOR_OUTPUT_WISE = 'output-wise'

class mlrl.boosting.config.parameters.RegressionLossParameter

Bases: NominalParameter

A parameter that allows to configure the loss function to be minimized during training in regression problems.

LOSS_SQUARED_ERROR_DECOMPOSABLE = 'squared-error-decomposable'

LOSS_SQUARED_ERROR_NON_DECOMPOSABLE = 'squared-error-non-decomposable'

class mlrl.boosting.config.parameters.ShrinkageParameter

Bases: FloatParameter

A parameter that allows to configure the shrinkage parameter, a.k.a. the learning rate, to be used.

class mlrl.boosting.config.parameters.StatisticFormatParameter

Bases: NominalParameter

A parameter that allows to configure the format to be used for the representation of gradients and Hessians.

STATISTIC_FORMAT_DENSE = 'dense'

STATISTIC_FORMAT_SPARSE = 'sparse'

class mlrl.boosting.config.parameters.StatisticTypeParameter

Bases: NominalParameter

A parameter that allows to configure the data type that should be used for representing gradients and Hessians.

STATISTIC_TYPE_FLOAT32 = '32-bit'

STATISTIC_TYPE_FLOAT64 = '64-bit'