This topic describes how to configure the Minimum Value and Maximum Value parameters for data features.

The Minimum Value and Maximum Value parameters for data features specify the lower limit and upper limit of the acceptable range of values for a metric. If a metric value is outside the specified range, the algorithm model determines that the metric is abnormal and generates an anomalous event. This anomalous event has the highest anomaly score 1.0, and an alert is triggered. If a metric value is within the specified range, the algorithm model automatically fits metric values and determines the distribution and change trends of the metric values. If the metric values abnormally fluctuate, the algorithm model generates an anomalous event.
Note If you cannot determine the acceptable range of values for a metric, you do not need to configure the Minimum Value or the Maximum Value parameter for data features of the metric. In this case, the algorithm model automatically fits metric values and determines the range of values for the metric. If you configure the range of values for a metric, the algorithm model can better determine the distribution of the metric values.
For example, the Minimum Value parameter for the cpu_usage metric is set to 0, and the Maximum Value parameter is set to 100. If you want to enable intelligent inspection for the cpu_usage metric and specify that an anomalous event is generated only when the value of the cpu_usage metric is outside the range of 0 to 10, you must set Feature to cpu_usage, Minimum Value to 0, Maximum Value to 100, and Time Series Segments to 10. After the configuration is complete, the algorithm model divides values from 0 to 100 into 10 buckets, and a data point falls into a bucket whose value range includes the data point. A data point specifies a value of the cpu_usage metric.
  • If a data point is 101, which is outside the range of 0 to 100, the algorithm model generates an anomalous event with an anomaly score of 1.0, and an alert is generated.
  • If the first data point is 3, the data point falls into the first bucket. If the second data point is 9, the data point also falls into the first bucket. If data points fall into the same bucket, the algorithm model considers the data points to be similar changes and does not generate anomalous events.
Important
  • If you specify invalid values for the Minimum Value and Maximum Value parameters for data features, alert storms can be triggered. If you cannot determine which values to specify, you can submit a ticket to consult.
  • If a metric value is outside the range that is specified by the Minimum Value and Maximum Value parameters, the algorithm model generates an anomalous event with an anomaly score of 1.0, and an alert is generated.
  • If a metric value is within the specified range, the algorithm model still captures abnormal time series changes. The value range of an anomaly score is (0, 1).
  • The intelligent inspection feature sends alert notifications based on the inhibition policy. Only one alert notification can be sent for a job per minute. The alerting system automatically aggregates all anomalous events that are captured within a minute. The system sends a notification only on the alert that is triggered by the anomalous event with the highest anomaly score. You can view the details about anomalous events on the dashboards of the project for which the intelligent inspection feature is enabled.