Used to generate a trendline for a periodic dataset. The key parameter period specifies the length of the period to be applied. This is used after a timechart, as in timechart(...) | holtwinters(1week). The function implements triple exponential smoothing, and adds an extra graph for every series in the graph. The computation assumes periodic data set (the period parameter specifies the length of the period), such as daily or weekly traffic measurements. The function works best when the timecharts span (width of each bucket) is a multiple of the period, as the smoothing is done in units of the buckets.

For example, a 30 day chart with a 7 day period would have 4 periods from day 2 to 30 (the first 2 days being ignored). Based on this, the first 3 periods (in this case period 1, 2 and 3) are fed into the smoothing/forecast computation, producing a "forecast" for the fourth period. Smoothing data for period 2, 3 and 4 are emitted, the last period being entirely computed.

ParameterTypeRequiredDefaultDescription
alphanumberfalse Data smoothing factor in the interval [0..1].
betanumberfalse Trend smoothing factor in the interval [0..1].
gammanumberfalse Seasonal change smoothing in the interval [0..1].
periodstringtrue Defines the trend's period, such as 1week or 1day |---| see the full list of the allowed time periods at Relative Time Syntax. [a]

[a] When you provide only one parameter, the implied parameter is period

Examples

Show event count graph with daily aggregates and compute the weekly trend.

humio
timechart(function=count(), span=1d) | holtwinters(period=7d)