Anomaly Analytic Unit

Basic Case

1. Create Analytic Unit
2. Choose detector type (Detector Type -> anomaly)
3. Click Create )
4. Enable Inspect mode to see HSR (Hastic Signal Representaion) -- data that comes to Hastic server Note: aggregation resolution (points on one time unit) of data depends on time range which is set when you create analytic unit
5. Related curves will be shown on the graph: HSR, Upper and Lower bounds (bounds are smoothed HSR plus/minus value of Confidence)
6. To change bounds of "normal" metric behavior you can change Confidence value. Click to field [5], change value (here we change confidence to 20% CPU usage) and click Save button [6].
7. Bounds are minimized and we got some "anomaly" data
8. Change Alpha value [7] to make bounds more or less smoothed and save unit [8] Note: Alpha takes value from 0 to 1
9. more "alpha" - more sharp bounds
10. less "alpha" - less sharp bounds
11. Click Detect button to get detections (Hastic will find segments which are out of bounds)

Seasonal data

1. You might have seasonality in your metric. For example, peaks every 24 hours
2. For our settings (Confidence: 0.2, Alpha: 0.3) this is anomaly
3. Click Seasonality [1] to label that it is normal behavior
4. Set season length to 24 hours [2 and 3] and save unit [4]
5. Click Label button [5] to label which segment of data is repeated
6. Label segment via Ctrl+Left Mouse
7. Click Detect button to show bounds with your repeated segment and new detections
8. Segment will repeated every 24h and become non-anomaly
9. Also, you can label another segment and it will be also repeated every 24h