Multi-period estimation function format and parameters -

Multi-period estimation functions estimate periods in time series data across different intervals. These functions extract periodic patterns using operations such as the Fourier transform.

Function list

Function	Description
ts_period_detect	Estimates the period of time series data over different time intervals.
ts_period_classify	Calculates the periodicity of a time series curve using the Fourier transform. Use this function to quickly determine if a curve has periodic patterns.

ts_period_detect

Syntax:

select ts_period_detect(x,y,minPeriod,maxPeriod)

The following table describes the parameters.

Parameter	Description	Value
x	The time column, sorted in ascending order.	A Unix timestamp in seconds.
y	The value column for each time point.	None
minPeriod	The estimated minimum period length, specified as a ratio of the total length of the time series.	A decimal value in the range (0, 1].
maxPeriod	The estimated maximum period length, specified as a ratio of the total length of the time series. Important The value of maxPeriod must be greater than minPeriod and cannot exceed 0.5. If you set maxPeriod to a value greater than 0.5, the system uses 0.5 by default.	A decimal value in the range (0, 1].

Example:

Query

* | select ts_period_detect(stamp, value, 0.2, 0.5) from ( select __time__ - (__time__ % 120) as stamp, avg(v) as value from log GROUP BY stamp order by stamp )

Output

The function returns an array that contains Unix timestamps, statistical values such as average traffic, and period status codes. A status code of 1.0 corresponds to a red circle in the following figure. You can visualize the output on a time series chart, as shown below.

In the following figure, the shaded area between two red circles represents one period. The curve patterns across different periods tend to be similar.

ts_period_classify

Syntax:

select ts_period_classify(stamp,value,instanceName)

The following table describes the parameters.

Parameter	Description	Value
stamp	The time column, sorted in ascending order.	A Unix timestamp in seconds.
value	The value column for each time point.	None
instanceName	The time series curve name.	None

Example:

Query

* and h : nu2h05202.nu8 | select ts_period_classify(stamp, value, name) from log

Output: The query returns a table that contains three columns: line_name (the timeline name, which is the Auto Scaling group ID), prob (the probability value), and type (the type flag, where -1.0 indicates an anomaly and 0.0 indicates normal). For example, the Auto Scaling group asg-2zgicin6zf5ewg188pg5 has a prob of 1.0 and a type of -1.0, which indicates that this timeline is identified as an anomaly.

The output includes the following columns:

Column	Description
line_name	The time series curve name.
prob	The proportion of the primary period within the time series curve. The value is in the range [0, 1]. A value of 0.15 is a typical experimental threshold.
type	The classification of the curve: type = -1: The curve length is too short (less than 64 points). type = -2: Indicates that the curve has a very high missing rate (the missing rate exceeds 20%). `type = 0.0`: The curve shows clear periodicity.