Penalise changepoints more

[multimeric]

Hi, thanks for writing this package.

I'm finding that EnvCpt finds a few more changepoints than I believe are actually present, so I'd like to penalise them a bit more. From looking at the documentation and source code it seems that envcpt passes all its ... args into cpt.reg (I'm looking at a regression changepoint model). However, even if I do envcpt(my_vector, pen.value=999999), don't notice any change in the resulting changepoints. Is there another way I'm supposed to do this?

Also, I think it would be helpful if pen.value were included in the manual for envcpt, since the function that it refers to (cpt.meanvar) is not exported.

Thanks.

[multimeric]

I now realise from looking at the manual file, that this is referring to the changepoint::cpt.meanvar. This cross-package reference doesn't show up in the PDF manual which is why I was confused. From looking at the changepoint manual it seems that I would first have to set penalty="manual", and then pen.value would work.

[rkillick]

Nope, your original inclination was correct. We haven't passed the pen.value argument to the cpt.reg function and have hard coded as MBIC penalty. I'll add this to the "enhancement" list but with the current backlog of work due to the pandemic I doubt I'll get to it before the end of the summer. I'd certainly welcome a pull request if you have time/inclination.

[multimeric]

I think it does seem to work, though:

data = c(1:100, 100:1, 1:100) + rnorm(300, sd = 5)
fit = EnvCpt::envcpt(data, models="trendcpt", penalty="Manual", pen.value=10)
plot(fit, colors = rep.int('red', 12))

versus

fit = EnvCpt::envcpt(data, models="trendcpt", penalty="Manual", pen.value=100)
plot(fit, colors = rep.int('red', 12))

[rkillick]

Ah yes, my past self must have changed this! This pandemic has ruined my memory of distinguishing what I intended to do and what I actually did! Thanks for trying it and seeing that it worked, saved me a job.