Skip to main content
Detect points where the underlying data distribution changes. Canary implements this by computing the energy gradient: the rate of change of the prediction error over a sliding window. This endpoint is fully compatible with the Azure Anomaly Detector changepoint API. Both /anomalydetector/v1.0/timeseries/changepoint/detect and /anomalydetector/v1.1/timeseries/changepoint/detect are supported.
​
array
required
Array of data points, each with timestamp (ISO 8601) and value (number). Minimum 12 points, maximum 8640 points. Timestamps must be sorted ascending with no duplicates.
​
string
default:"none"
Time interval between points. One of: yearly, monthly, weekly, daily, hourly, minutely, secondly, microsecond, none.
​
integer
default:1
Multiplier for granularity.
​
integer
default:0
Seasonality period. Set to 0 for auto-detection.
​
number
Maximum fraction of points flagged. Must be between 0 and 0.5 (exclusive).
​
integer
default:95
Detection sensitivity (0-99).
​
integer
Detected seasonality period.
​
boolean[]
Whether each point is a changepoint.
​
number[]
Confidence score from 0.0 to 1.0 for each point.

​
How Changepoint Detection Works

  1. The energy gradient is computed: gradient[i] = mean(energies[i:i+k]) - mean(energies[i-k:i]) where k = max(period, 5).
  2. A point is a changepoint if its gradient exceeds the 95th percentile of all gradients.
  3. The gradient must be a local maximum (greater than or equal to both neighbors).
  4. Minimum spacing between changepoints equals the detected period.