At 02:27 PM 11/15/2009 -0500, Terry Reedy wrote:
Sridhar Ratnakumar wrote:
Yes. If PyPI is going to have release-specific ratings, I suggest
that each PyPI page (distname-version) show two rating bars:
a) release rating (logged-in user editable)
b) overall rating (averaged from all release ratings)
I would make the second an average over all past releases
I would also suggest making it a so-called exponentially weighted
average. Assume that past ratings are frozen and the individual
votes tossed, the update when a new release arrives would be something like
past = .8 * past + .2 * current # adjust weights to taste
current = <undefined>
This way, a blunder release, once corrected, would have limited
effect on the overall average and its effect would disappear with time.
You need to weight also for the number of ratings, not just the overall rating.
_______________________________________________
Catalog-SIG mailing list
[email protected]
http://mail.python.org/mailman/listinfo/catalog-sig
_______________________________________________
Catalog-SIG mailing list
[email protected]
http://mail.python.org/mailman/listinfo/catalog-sig
