Guys,
when I need you on skype or gmail for serious reasons (and not to talk about bullshit) you're never there...
Anyway... the question is the following: do you know how to compute the smoothness of a curve in R (and if this is possible)? I want to smooth a time series (I'm going to use loess) and then compute how much smooth is the curve that I ended up with; the idea is to compute some smoothness indicator, like, as an example, the penalty term that you consider while fitting a spline (\int |f^{''}(t)|^2 dt )... Any suggestion (I can't understand if the R loess function gives as output some usefull quantities or not...), have you ever dealt with this?
Thank you!!!
Ciao bei e bea
Checca
P.S.: please (toni) no comments about that fact that I permute, or at least find some new funny sentences, lately you were using always the same boring ones...
Subscribe to:
Post Comments (Atom)
Anyway... are you all in holidays??? this blog looks like the department of statistics on july and august of the last summer: empty, nobody passing by... (bustra, do you remember it... no comment...)
ReplyDelete...I need somebody,
ReplyDeleteHelp, not just anybody,
Help, you know I need someone, help.
R loess uses a standard smoothing parameter, hence if you have two different outcome and you want to fit different smoothing curves like
ReplyDeletey1 = read.quel.che.te.voi()
y2 = read.quel.che.te.voi()
x = read.quel.che.te.voi()
fit1 = loess(y1~x)
fit2 = loess(y2~x)
fit1$par$span
fit1$par$span
you'll see 0.75 (default choice). Than you can obtain smoother curve changing this value in the formula.
But I thik thsat is something relative to your data and so is not a measure of smoothness.
Your idea is smart but its difficult to compute.
My suggestion is take a grid and approximate your curve with a polygonal (una spezzata) and compute something like
sum_{i =1}^{ngrid} slope_i^2
make sense?
for non-paramtric stuff usually the measure of smoothness is computed via the equivalent number of degrees of freedom. I don't know if the loess function allow you to extract this information though. For sure you can do it for splines
ReplyDeleteindeed your idea is good but it would be data related, so not a global measure...
Thank you! also bruno suggested to use the equivalent degrees of freedom, and the loess function in R should have them in the output. bruno also suggested me a way to compute the integral via incremental ratio (rapporto incrementale, is this the right translation?) twice to compute the second derivative and then numerical appriximation of the integral.
ReplyDeleteNow the point is that I really don't know wich curve I'm interested in... I know, sounds wired... @Ila: I need you and two omer beers one of the next evenings... See you tonight! hihihi super surprise hihihi
Thanks tony!
Ciao bei!
Checca
Checca you won't believe it but there is another Professor who teaches permutation tests! It's Philip B. Stark do you know him?
ReplyDeleteThe question is for you too, Monjed!
Noooo!!! cool!!!!!!!!!!!!! no, I don't know him... if you talk with him... ask him if they're looking for somebody to work on permutation methods in berkley!
ReplyDeletePermutation power!!!
Monjed, too late, if they need only one person... the position is mine, I asked for it before you!
ReplyDeleteChecca nn so niente di niente di quello che hai scritto ma ti voglio bene:)
ReplyDeleteDear Maestro,
ReplyDeleteI do not know him, I just read about him, he looks a great one.
I think could be a good reference. But I am not interested to go USA, I am interested in European Doctorate certificate, so I would like to have one from Europe.
Thanks all