Configuration of how features in Featurestore are monitored.
JSON representation |
---|
{ "snapshotAnalysis": { object ( |
Fields | |
---|---|
snapshotAnalysis |
The config for Snapshot Analysis Based feature Monitoring. |
importFeaturesAnalysis |
The config for ImportFeatures Analysis Based feature Monitoring. |
numericalThresholdConfig |
Threshold for numerical features of anomaly detection. This is shared by all objectives of Featurestore Monitoring for numerical features (i.e. Features with type ( |
categoricalThresholdConfig |
Threshold for categorical features of anomaly detection. This is shared by all types of Featurestore Monitoring for categorical features (i.e. Features with type ( |
SnapshotAnalysis
Configuration of the Featurestore's Snapshot Analysis Based Monitoring. This type of analysis generates statistics for each feature based on a snapshot of the latest feature value of each entities every monitoringInterval.
JSON representation |
---|
{ "disabled": boolean, "monitoringInterval": string, "monitoringIntervalDays": integer, "stalenessDays": integer } |
Fields | |
---|---|
disabled |
The monitoring schedule for snapshot analysis. For EntityType-level config: unset / disabled = true indicates disabled by default for Features under it; otherwise by default enable snapshot analysis monitoring with monitoringInterval for Features under it. feature-level config: disabled = true indicates disabled regardless of the EntityType-level config; unset monitoringInterval indicates going with EntityType-level config; otherwise run snapshot analysis monitoring with monitoringInterval regardless of the EntityType-level config. Explicitly Disable the snapshot analysis based monitoring. |
monitoringInterval |
Configuration of the snapshot analysis based monitoring pipeline running interval. The value is rolled up to full day. If both A duration in seconds with up to nine fractional digits, ending with ' |
monitoringIntervalDays |
Configuration of the snapshot analysis based monitoring pipeline running interval. The value indicates number of days. |
stalenessDays |
Customized export features time window for snapshot analysis. Unit is one day. Default value is 3 weeks. Minimum value is 1 day. Maximum value is 4000 days. |
ImportFeaturesAnalysis
Configuration of the Featurestore's ImportFeature Analysis Based Monitoring. This type of analysis generates statistics for values of each feature imported by every ImportFeatureValues
operation.
JSON representation |
---|
{ "state": enum ( |
Fields | |
---|---|
state |
Whether to enable / disable / inherite default hebavior for import features analysis. |
anomalyDetectionBaseline |
The baseline used to do anomaly detection for the statistics generated by import features analysis. |
State
The state defines whether to enable ImportFeature analysis.
Enums | |
---|---|
STATE_UNSPECIFIED |
Should not be used. |
DEFAULT |
The default behavior of whether to enable the monitoring. EntityType-level config: disabled. feature-level config: inherited from the configuration of EntityType this feature belongs to. |
ENABLED |
Explicitly enables import features analysis. EntityType-level config: by default enables import features analysis for all Features under it. feature-level config: enables import features analysis regardless of the EntityType-level config. |
DISABLED |
Explicitly disables import features analysis. EntityType-level config: by default disables import features analysis for all Features under it. feature-level config: disables import features analysis regardless of the EntityType-level config. |
Baseline
Defines the baseline to do anomaly detection for feature values imported by each ImportFeatureValues
operation.
Enums | |
---|---|
BASELINE_UNSPECIFIED |
Should not be used. |
LATEST_STATS |
Choose the later one statistics generated by either most recent snapshot analysis or previous import features analysis. If non of them exists, skip anomaly detection and only generate a statistics. |
MOST_RECENT_SNAPSHOT_STATS |
Use the statistics generated by the most recent snapshot analysis if exists. |
PREVIOUS_IMPORT_FEATURES_STATS |
Use the statistics generated by the previous import features analysis if exists. |
ThresholdConfig
The config for Featurestore Monitoring threshold.
JSON representation |
---|
{ // Union field |
Fields | |
---|---|
Union field
|
|
value |
Specify a threshold value that can trigger the alert. 1. For categorical feature, the distribution distance is calculated by L-inifinity norm. 2. For numerical feature, the distribution distance is calculated by Jensen–Shannon divergence. Each feature must have a non-zero threshold if they need to be monitored. Otherwise no alert will be triggered for that feature. |