After a monitor has run for some time, you might want to adjust its detection settings and train it with different periods of data. Sometimes you might need to adjust a new monitor several times to ensure it's assessing your data quality correctly. Or perhaps a shift in data patterns has caused a live monitor to begin detecting more incidents than it should. You can easily edit a monitor to adjust its behavior accordingly.
- In the metric chart, select Monitor -> Edit monitor -> your monitor's name.
- You can't edit a live monitor, so at the top right corner, select Live, then select Turn offline in the pop-up.
- You can rename the monitor on the Define tab, but most of the editable settings are on the Train tab.
- Make any changes in the Training and Detection settings sections. If desired, toggle Incident learning on (right) or off (left). Note that you can pick several training periods; this can help you to exclude periods that have anomalous data. Also note that if your data asset has slices you can choose which slice to train and review output for. Select Save and train to begin training using the periods you specified.
Training might take a while. If it's taking too long, try using a shorter training period, or if your metric has slices try training with a different slice.
- After training completes, preview the revised monitor to make sure your changes had the desired effect. On the Preview tab, pick a date for Start and End, then select Generate preview.
- If you're not satisfied with the previewed changes, make adjustments on the Train tab, select Save and train, and then generate a new preview. Repeat until the monitor is behaving as desired.
- At the top right, select Offline to make the monitor live again.
Updated 29 days ago