Share via


LbfgsTrainerBase<TOptions,TTransformer,TModel>.OptionsBase Class

Definition

Base options class for trainer estimators deriving fromLbfgsTrainerBase<TOptions,TTransformer,TModel>.

public abstract class LbfgsTrainerBase<TOptions,TTransformer,TModel>.OptionsBase : Microsoft.ML.Trainers.TrainerInputBaseWithWeight where TOptions : LbfgsTrainerBase<TOptions,TTransformer,TModel>.OptionsBase, new() where TTransformer : ISingleFeaturePredictionTransformer<TModel> where TModel : class
type LbfgsTrainerBase<'Options, 'ransformer, 'Model (requires 'Options :> LbfgsTrainerBase<'Options, 'ransformer, 'Model>.OptionsBase and 'Options : (new : unit -> 'Options) and 'ransformer :> ISingleFeaturePredictionTransformer<'Model> and 'Model : null)>.OptionsBase = class
    inherit TrainerInputBaseWithWeight
Public MustInherit Class LbfgsTrainerBase(Of TOptions, TTransformer, TModel).OptionsBase
Inherits TrainerInputBaseWithWeight

Type Parameters

TOptions
TTransformer
TModel
Inheritance
LbfgsTrainerBase<TOptions,TTransformer,TModel>.OptionsBase
Derived

Constructors

LbfgsTrainerBase<TOptions,TTransformer,TModel>.OptionsBase()

Fields

DenseOptimizer

Force densification of the internal optimization vectors. Default is false.

EnforceNonNegativity

Enforce non-negative weights. Default is false.

ExampleWeightColumnName

Column to use for example weight.

(Inherited from TrainerInputBaseWithWeight)
FeatureColumnName

Column to use for features.

(Inherited from TrainerInputBase)
HistorySize

Number of previous iterations to remember for estimating the Hessian. Lower values mean faster but less accurate estimates.

InitialWeightsDiameter

Initial weights scale.

L1Regularization

L1 regularization weight.

L2Regularization

L2 regularization weight.

LabelColumnName

Column to use for labels.

(Inherited from TrainerInputBaseWithLabel)
MaximumNumberOfIterations

Number of iterations.

NumberOfThreads

Number of threads. Null means use the number of processors.

OptimizationTolerance

Tolerance parameter for optimization convergence. (Low = slower, more accurate).

Quiet

Determines whether to produce output during training or not.

StochasticGradientDescentInitilaizationTolerance

Run SGD to initialize LR weights, converging to this tolerance.

Applies to