Single change

Detecting change point is a two step process:

1. Is there a change point?

2. When did the change happen?

Is there a change

mcppv uses randomization to calculate the p-value of observing the particular sequence. It uses shuffle! to re-arrange the centered time series. For repeatable results, set the random number seed first.

using ChangePointMean
using Random

ts = [1, 2, 1, 2, 1, 2, 10, 11, 10, 11, 10]
for _ in 1:10
    Random.seed!(888)
    println(mcppv(ts))
end

0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809
0.005988023952095809

If the random number seed is not set, the results will not be repeatable:

ts = [1, 2, 1, 2, 1, 2, 10, 11, 10, 11, 10]
Random.seed!(888)
for _ in 1:10
    println(mcppv(ts))
end

0.005988023952095809
0.005988023952095809
0.005988023952095809
0.001996007984031936
0.005988023952095809
0.003992015968063872
0.001996007984031936
0.003992015968063872
0.005988023952095809
0.003992015968063872

When is the change

mcptime estimates the location of the change point by finding the location of minimum sum of squares when the time series is split into two halves.

mcptime(ts)

7

A single change point

mcpoint uses mcppv and mcptime together to return the location of a change point, the start of a new shifted sub-series:

mcpoint(ts)

7

If there is no change point, mcpoint returns 0:

mcpoint([1, 1, 1, 1, 1, 1, 1, 1, 1])

0

You can use minlen to control the length of segments:

mcptime([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 10, 10, 10])

12

mcptime([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 10, 10, 10]; minlen = 4)

11