Share via


AutoMLExperimentExtension.SetSmacTuner Method

Definition

Set Microsoft.ML.AutoML.SmacTuner as tuner for hyper-parameter optimization. The performance of smac is in a large extend determined by numberOfTrees, nMinForSpit and splitRatio, which are used to fit smac's inner regressor.

public static Microsoft.ML.AutoML.AutoMLExperiment SetSmacTuner (this Microsoft.ML.AutoML.AutoMLExperiment experiment, int numberInitialPopulation = 20, int fitModelEveryNTrials = 10, int numberOfTrees = 10, int nMinForSpit = 2, float splitRatio = 0.8, int localSearchParentCount = 5, int numRandomEISearchConfigurations = 5000, double epsilon = 1E-05, int numNeighboursForNumericalParams = 4);
static member SetSmacTuner : Microsoft.ML.AutoML.AutoMLExperiment * int * int * int * int * single * int * int * double * int -> Microsoft.ML.AutoML.AutoMLExperiment
<Extension()>
Public Function SetSmacTuner (experiment As AutoMLExperiment, Optional numberInitialPopulation As Integer = 20, Optional fitModelEveryNTrials As Integer = 10, Optional numberOfTrees As Integer = 10, Optional nMinForSpit As Integer = 2, Optional splitRatio As Single = 0.8, Optional localSearchParentCount As Integer = 5, Optional numRandomEISearchConfigurations As Integer = 5000, Optional epsilon As Double = 1E-05, Optional numNeighboursForNumericalParams As Integer = 4) As AutoMLExperiment

Parameters

numberInitialPopulation
Int32

Number of points to use for random initialization.

fitModelEveryNTrials
Int32

re-fit random forests in smac for every N trials.

numberOfTrees
Int32

number of regression trees when fitting random forest.

nMinForSpit
Int32

minimum number of data points required to be in a node if it is to be split further for fitting random forest in smac.

splitRatio
Single

split ratio for fitting random forest in smac.

localSearchParentCount
Int32

Number of search parents to use for local search in maximizing EI acquisition function.

numRandomEISearchConfigurations
Int32

Number of random configurations when maximizing EI acquisition function.

epsilon
Double

the threshold to exit during maximizing EI acquisition function.

numNeighboursForNumericalParams
Int32

Number of neighbours to sample from when applying one-step mutation for generating new parameters.

Returns

Applies to