sapsi
1
Let me explain. Suppose i have a model with
y ~ s(time,k=4)
where time = 1 ... N
and i want the spline to be insenstive to changes in y
when time \in [N-14,N]
(here N
is ~700).
One thing i’ve been doing is placing tight priors (to reduce the prior sigma for the spline)
prior = c( set_prior("exponential(8)", class = "sds",coef="s(i,k=4)"))
but how can i allow it be ‘freer’ for time<(N-14)
but very insensitive for N >= time >=(N-14)
It sounds like you want a restricted cubic spline. If you search for rcs
on Discourse a few threads will come up about how to do that.
sapsi
3
Thanks for the suggestion. Will look into it
There are splines where you can specify linear effects at the end, or constant effect.
You could set the time N >= time >=(N-14) to time - 14.
You may also get some ideas from wavelet implementations depending on your signal.
https://www.mathworks.com/help/wavelet/ug/dealing-with-border-distortion.html
sapsi
5
Sorry for the late reply. Yes this is very nice idea.