rkillick/changepoint

Segment check or residual check?

Closed this issue · 4 comments

Jiangchao3 commented

Hi, killick

I have completed detecting changepoints for my data. Following your guide, I should do the assumption check according to this website http://members.cbio.mines-paristech.fr/~thocking/change-tutorial/RK-CptWorkshop.html#checking-assumptions-if-time-allows.

You provide two methods, one is segment check and the other is residual check. Any differences between these two checks? I want to know which one should I choose? Or should I do both these two checks?

Thanks!
Jiangchao

Jiangchao3 commented

I noticed that the segment check is applying the Shapiro-Wilk normality test, KS test, qqPlot, AcfPlot to each segment. It means I should test or plot N+1 times if there are N changepoints. The residual check just test and plot once. I think I should choose the residual check because my data have more than 10 changepoints.

rkillick commented

Personally, I would always use the residual check at this gives larger n and thus more power. I have removed the segment check in the latest version of my course.

Jiangchao3 commented

Thank you! I have noticed the change in your latest course.