This page shows the results of running the anomaly detection function on some test data for some anomaly threshold configuration.
{% if plot_data and lookup %}The test data series is shown in grey. The segment and segment endpoints are shown in orange; if a revision is specified, a blue line is also shown indicating the position of the revision. Above each detected anomaly is a label showing the revision number and percent change from the median before the anomaly to the median after the anomaly.
{% endif %} {% if error %}{{error}}
{% endif %}About the threshold parameters
-
max_window_size
is the number of points that's analyzed at once. A larger window size means that more points before a change-point can be considered. The window is split into two segments. Must be at least twice as big as themin_segment_size
. -
min_segment_size
determines how many points must be on both sides of the change-point. A larger value makes it less sensitive to noise, but it also increases the time before an alert is fired. -
min_relative_change
is the minimum relative difference between the median of the points before and the median of the points after. that's reported. For example, to disallow alerts for a change of less than 5%, this can be set to 0.05. -
min_absolute_change
sets a threshold on the absolute difference before and after. -
multiple_of_std_dev
is a threshold on the jump size measured in multiples of the standard deviation of the points. For example, a value of 3.0 means that only a jump 3 or more times larger than a sample of points around the change point is considered. -
min_steppiness
is a measure of how similar the change is to a perfect step, where 0.0 corresponds to a flat line or completely random noise, and 1.0 corresponds to a any perfect step.
Recent Anomaly entities in the datastore for this test:
End Revision | Bug ID | Median before anomaly | Median after anomaly | Percent changed | Timestamp |
---|---|---|---|---|---|
{{alert.revision}} | {{alert.bug_id}} | {{alert.median_before}} | {{alert.median_after}} | {{alert.percent_changed}} | {{alert.timestamp}} |