|
Description |
An exponential moving average is similar to a moving average, except that more weight is given to the latest data. It is also known as an exponentially weighted moving average. |
|
|
Input |
Time Series: Multiple (Single or multiple time-series) |
|
|
Parameters |
Period |
Mandatory. Relative_time (Minutes, Hours, Days) (default=10 days) |
|
Output |
Transformed Time Series - multiple |
|
|
Available in |
Alerts + Composite + Dashboards |
|
Usage example
The expMovingAverage function is useful for smoothing noisy metrics.
Take the following raw metric for example, which counts the number of application errors in a given instance:
It is a sparse metric, as the majority of the input here is only zeros. In addition, the seasonality is not perfect, as the daily peaks emerge with some jitter. In fact, this noise hampers the baseline model selection, and leaves this metric inadequate for anomaly detection (all those daily surges will be falsely classified as anomalies).
After applying the expMovingAverage function (using a 45-minute moving window), the baseline for this metric becomes more focused:
Note: the expMovingAverage smoothes the original time series, typically reducing the magnitude of the spikes. Here is the same spike, this time using a 60 minutes moving window:
