Today I learnt, via Sam Power from Bristol, that the legendary IJ Good and the possibly legendary (I really don’t know) RA Gaskins suggested, in their 1971 paper on density estimation, referring to the roughness penalty from density estimation (or non-linear regression) as the flamboyance1 functional. And it would be a crime if we, as a field, did not take up this terminology.
Don’t say: Splines balance the bias and the variance.
Do say: Splines support flamboyance, but only if it’s repeatable.
Footnotes:
1 They specify “flamboyant in the sense of having a wavy edge”. It was the 70s after all. Being wavy was a big thing.
Now that’s going back to our roots.
Any doogians still out there?
https://www.oreilly.com/library/view/bayesian-statistics-an/9781118359778/OEBPS/c04-sec1-0006.htm
Better yet, the original Good (Nature, 1971) flamboyancy penalty was … Fisher information! int (sqrt(f(x))’)^2)dx this was deemed “too straight’ in subsequent work with Gaskins, and they consider quadratic penalties on second derivatives. Sliverman’s (1981) proposal to penalize int ((log f)”’)^2 dx was better since it shrinks toward the Gaussian, if you like Gaussians. If not:
http://www.econ.uiuc.edu/~roger/research/densiles/Doksum.pdf
I like the word. But but.
I might also add that some countries such as Finland are not really liberal. Here you are gay or whatever already when wearing a wrong kind of a shirt. And god forbid if you like the arts. Because nothing could be more gay than something like poetry, right?
There is also something peculiarly Finnish in these defamation businesses. No one dares to ask anything face-to-face. The mob decides according to its fantasies.
Now back to writing a paper on propaganda. Splines here, splines there.