FederatedDriftOptions
Options & Configuration · AiDotNet.Models.Options
Configuration options for federated concept drift detection and adaptation.
For Beginners
In production, client data changes over time (e.g., fraud patterns evolve, user behavior shifts seasonally). These options configure how the FL system detects when client data has drifted from the training distribution and how it adapts.
Constructors
| Constructor | Summary |
|---|---|
FederatedDriftOptions() |
Properties
| Property | Summary |
|---|---|
FederatedDriftMethod Method { get; set; } |
Gets or sets the drift detection method. |
bool AdaptAggregationWeights { get; set; } |
Gets or sets whether to automatically adapt aggregation weights based on drift scores. |
bool Enabled { get; set; } |
Gets or sets whether drift detection is enabled. |
bool TriggerSelectiveRetraining { get; set; } |
Gets or sets whether to trigger selective retraining when global drift is detected. |
double GlobalDriftThreshold { get; set; } |
Gets or sets the global drift threshold above which all clients are notified. |
double MinDriftWeight { get; set; } |
Gets or sets the minimum weight multiplier for heavily drifting clients. |
double SensitivityThreshold { get; set; } |
Gets or sets the sensitivity threshold for drift detection. |
int DetectionFrequency { get; set; } |
Gets or sets how often to run drift detection (in rounds). |
int LookbackWindowRounds { get; set; } |
Gets or sets the number of recent rounds to consider for drift analysis. |