|Problems? Is your data what you think it is?|
Re: Hockey Sticksby chromatic (Archbishop)
|on Jan 17, 2012 at 01:06 UTC||Need Help??|
That's... not exactly what I understand statistics to mean.
Way back in the day when Rakudo had graphs of spec test data every day, Solomon Foster spent a lot of time writing hundreds (thousands?) of tests for complex numerical operations. He put a lot of energy into a thankless task and deserves a lot of credit for that, but the best-kind-of-true fact that "Rakudo passes 10% more tests this month than last!" didn't translate into a Rakudo that was better for most people to use for serious purposes.
Similarly the fact that Niezca passes more tests than Rakudo doesn't necessarily mean that Niezca is more usable or useful or complete than Rakudo. It's not even clear if the tests passed are the right tests or are meaningful to its utility. (Besides, given Rakudo's rewrite-induced regressions, it shouldn't be surprising that another implementation can overtake it.)
Further, if you want to talk about trends and acceleration, you probably need more than a month's worth of data points to do so. You can't get much of a meaningful derivative from a graph with so much noise; you need to normalize it, and you need to normalize it to a normative period. Otherwise, fun with statistics means that we can predict that, based on the spectest measurements of one platform between mid-December 2011 and mid-January 2012, we expect:
"the awkward situation with Rakudo Star has helped obscure the fantastic good news in Perl 6."
I hear that other than that, Mary Todd Lincoln rather liked the play.
Improve your skills with Modern Perl: the free book.