e.divisve produced breakpoints depending on min.size

[istvan60]

Dear Authors!

I was wondering, what could be the problem, when running e.divisive on multiple time series no matter how the function is parametrized the step between the change points equals approx. the min.size.

If min.size=100 then the gap between the breakpoints will scatter around 100, and so on.

library(ecp)
library(geoChronR) #only needed for plotting

start_time=1779 #the actual date of the first data point
res_short=e.divisive(as.matrix(df.wide_short), sig.lvl=0.001, R=1000, min.size=10, alpha=1)
corr_estimates_short=res_short$estimates+start_time
plotTimeseriesStack(plot.df_short_filtered)+
    geom_vline(xintercept = corr_estimates_short)

Any assistance is highly appreciated!

Yours sincerely,

István

[zwenyu]

It's likely because the time series is not piece-wise stationary with the specified min.size so the algorithm will try to segment as much as possible. You may want to try transforming the time series first.