Yeah. Centroid (and spectral stuff) would benefit from initial log/log computation (I remember @tremblap being super amped about that being on the horizon a couple years ago now
I nudge about so many things it’s hard for me to keep track of them…
On that note, I presume this would behave in a similar way to peek~, but going into gen doesn’t shit the threading bed (ala js) does it? As in, does it go into the low priority queue when being processed inside gen?
IIRC (which I might well not) the pause has been partly due to figuring out what the higher moments would mean / should be approached in the log case. The centroid is easy enough to interpret as a weighted geometric mean of the linear stuff, but the standard formulations for spectral spread etc. make a little bit less sense in that context.