AnomalyDetectionCatalog 類別

定義

用來 MLContext 建立異常偵測元件的實例的類別,例如定型器和評估工具。

public sealed class AnomalyDetectionCatalog : Microsoft.ML.TrainCatalogBase
type AnomalyDetectionCatalog = class
    inherit TrainCatalogBase
Public NotInheritable Class AnomalyDetectionCatalog
Inherits TrainCatalogBase
繼承
AnomalyDetectionCatalog

屬性

Trainers

異常偵測的定型器清單。

方法

ChangeModelThreshold<TModel>(AnomalyPredictionTransformer<TModel>, Single)

使用指定的 AnomalyPredictionTransformer<TModel> 建立新 threshold。 如果提供的 threshold 與閾值相同, model 它只會傳 model 回 。 請注意,閾值預設為 0.5,有效分數範圍從 0 到 1。

Evaluate(IDataView, String, String, String, Int32)

評估評分的異常偵測資料。

擴充方法

DetectEntireAnomalyBySrCnn(AnomalyDetectionCatalog, IDataView, String, String, SrCnnEntireAnomalyDetectorOptions)

建立 Microsoft.ML.TimeSeries.SrCnnEntireAnomalyDetector ,它會使用 SRCNN 演算法偵測整個輸入的逾時異常狀況。

DetectEntireAnomalyBySrCnn(AnomalyDetectionCatalog, IDataView, String, String, Double, Int32, Double, SrCnnDetectMode)

建立 Microsoft.ML.TimeSeries.SrCnnEntireAnomalyDetector ,它會使用 SRCNN 演算法偵測整個輸入的逾時異常狀況。

DetectSeasonality(AnomalyDetectionCatalog, IDataView, String, Int32, Double)

在時間序列資料中,季節性 (或週期性) 是特定定期發生的變化,例如每週、每月或每季。

這個方法會採用四位分析技術,以偵測這個可預測的間隔 (或期間) 。 假設輸入值有相同的時間間隔 (例如,依時間戳記) 排序每秒所收集的感應器資料,此方法會採用時間序列資料的清單,並傳回輸入季節性資料的一般期間,如果可預測的波動或模式可在整個輸入值期間內遞迴或重複。

如果找不到這類模式,則傳回 -1,也就是說,輸入值不會遵循季節性波動。

LocalizeRootCause(AnomalyDetectionCatalog, RootCauseLocalizationInput, Double, Double)

建立 RootCause ,以使用決策樹演算法將根本原因當地語系化。

LocalizeRootCauses(AnomalyDetectionCatalog, RootCauseLocalizationInput, Double, Double)

輸出 的 RootCause 已排序清單。 順序對應到最可能成為根本原因的備妥原因。

適用於