FastTreeTweedieTrainer 類別

定義

IEstimator<TTransformer> 用於使用 Tweedie 損失函式來定型決策樹回歸模型。 此定型器是 Poisson、複合 Poisson 和 gamma 回歸的一般化。

public sealed class FastTreeTweedieTrainer : Microsoft.ML.Trainers.FastTree.BoostingFastTreeTrainerBase<Microsoft.ML.Trainers.FastTree.FastTreeTweedieTrainer.Options,Microsoft.ML.Data.RegressionPredictionTransformer<Microsoft.ML.Trainers.FastTree.FastTreeTweedieModelParameters>,Microsoft.ML.Trainers.FastTree.FastTreeTweedieModelParameters>
type FastTreeTweedieTrainer = class
    inherit BoostingFastTreeTrainerBase<FastTreeTweedieTrainer.Options, RegressionPredictionTransformer<FastTreeTweedieModelParameters>, FastTreeTweedieModelParameters>
Public NotInheritable Class FastTreeTweedieTrainer
Inherits BoostingFastTreeTrainerBase(Of FastTreeTweedieTrainer.Options, RegressionPredictionTransformer(Of FastTreeTweedieModelParameters), FastTreeTweedieModelParameters)
繼承

備註

若要建立此定型器,請使用FastTreeTweedie 或 FastTreeTweedie (Options)

輸入和輸出資料行

輸入標籤資料行資料必須是 Single。 輸入特徵資料行資料必須是 的已知大小向量 Single

此定型器會輸出下列資料行:

輸出資料行名稱 資料行類型 描述
Score Single 模型所預測的未系結分數。

定型器特性

機器學習工作 迴歸
是否需要正規化?
是否需要快取?
除了 Microsoft.ML 之外,還需要 NuGet Microsoft.ML.FastTree
可匯出至 ONNX

定型演算法詳細資料

Tweedie 提升模型遵循從 Yang、Quan 和 Zou 的漸層Tree-Boosted Tweedie 複合 Poisson 模型所建立的保險進階預測中建立的數學。 如需漸層提升的簡介和詳細資訊,請參閱 :Wikipedia:漸層提升 ( 漸層樹狀結構提升) 或 窮盡函式近似值:漸層提升機器

如需使用範例的連結,請參閱另請參閱一節。

欄位

FeatureColumn

定型器預期的功能資料行。

(繼承來源 TrainerEstimatorBase<TTransformer,TModel>)
GroupIdColumn

排名定型器預期的選擇性 groupID 資料行。

(繼承來源 TrainerEstimatorBaseWithGroupId<TTransformer,TModel>)
LabelColumn

定型器預期的標籤資料行。 可以是 null ,表示標籤不會用於定型。

(繼承來源 TrainerEstimatorBase<TTransformer,TModel>)
WeightColumn

定型器預期的加權資料行。 可以是 null ,表示加權不會用於定型。

(繼承來源 TrainerEstimatorBase<TTransformer,TModel>)

屬性

Info

IEstimator<TTransformer> 用於使用 Tweedie 損失函式來定型決策樹回歸模型。 此定型器是 Poisson、複合 Poisson 和 gamma 回歸的一般化。

(繼承來源 FastTreeTrainerBase<TOptions,TTransformer,TModel>)

方法

Fit(IDataView)

定型並傳 ITransformer 回 。

(繼承來源 TrainerEstimatorBase<TTransformer,TModel>)
Fit(IDataView, IDataView)

FastTreeTweedieTrainer使用定型和驗證資料來定型 , 會 RegressionPredictionTransformer<TModel> 傳回 。

GetOutputSchema(SchemaShape)

IEstimator<TTransformer> 用於使用 Tweedie 損失函式來定型決策樹回歸模型。 此定型器是 Poisson、複合 Poisson 和 gamma 回歸的一般化。

(繼承來源 TrainerEstimatorBase<TTransformer,TModel>)

擴充方法

AppendCacheCheckpoint<TTrans>(IEstimator<TTrans>, IHostEnvironment)

將「快取檢查點」附加至估算器鏈結。 這可確保下游估算器會針對快取的資料定型。 在接受多個資料通過的定型器之前,先有快取檢查點會很有説明。

WithOnFitDelegate<TTransformer>(IEstimator<TTransformer>, Action<TTransformer>)

指定估算器時,傳回包裝物件,一旦 Fit(IDataView) 呼叫委派即可。 估算器通常務必傳回符合專案的相關資訊,這就是 Fit(IDataView) 方法傳回特定型別物件的原因,而不是只傳回一般 ITransformer 。 不過,同時, IEstimator<TTransformer> 通常會形成具有許多物件的管線,因此我們可能需要透過建置估算器的鏈結,而我們想要取得轉換器的估算器 EstimatorChain<TLastTransformer> 會在此鏈結的某處放置。 在該案例中,我們可以透過這個方法附加呼叫一次會呼叫的委派。

適用於

另請參閱