>>>>> On Wed, 2 Dec 2009 14:02:27 +0000, Barbie <bar...@missbarbell.co.uk> said:
> Hi Andreas, > BTW I think the site could make very interesting reading, however, I > suggest you add something on the about page to explain what the columns > mean for the regression tables. > I have no ideas what the 3 number columns represent, and the R^2, N and > K are meaningless to me I'm afraid. I suspect they have mathematical > meanings, but I don't have that high level background, and suspect there > are others who would be as mystified. If you can include references and > links (Wikipedia maybe?) that would help. Sorry for that. I beefed the "about" page up a little and added column names to all tables. I was too deep into the matter to think of the trivial things. I still think this needs a few iterations of improvement. You can read the diff at http://repo.or.cz/w/andk-cpan-tools.git/commitdiff/f07c51aef089e40b77e5cfd80eb85d95471eb3cb >>>>> On Wed, 02 Dec 2009 22:23:52 +0100, David Landgren <da...@landgren.net> >>>>> said: > I think I understand what this is telling me, but there's a lot of > information and it's difficult to pin-point what is important. Looking > at the File-Path distro, with 300+ passes to 3 or so fails, I can see > what appears to influence the likelihood of a fail. It's really just like a grep over all results without downloading them. Next step for you is to click on the title of the Regression qr:(Failed test\s+\S+.*) table and get links to the individual tests. > Then again: looking at BSD-Process, which will fail anywhere except on > FreeBSD, I can't see where it highlights the fact that, well hey, this > module tends to fail with surprising regularity on Linux. I reprint a few lines from the Regression conf:archame+osvers table: Name Theta StdErr T-test [0='const'] 0.0000 0.0888 0.00 [2='eq_amd64-freebsd 6.2-release'] 0.6667 0.1356 4.92 [7='eq_i386-freebsd 6.1-release'] 0.5000 0.1538 3.25 [12='eq_i386-freebsd-64int 6.0-stable'] 1.0000 0.1985 5.04 [13='eq_i386-freebsd-64int 6.1-release'] 1.0000 0.1985 5.04 [14='eq_i386-freebsd-64int 7.2-release'] 0.0000 0.1538 0.00 [15='eq_i386-freebsd-thread-multi 6.1-release'] 1.0000 0.1985 5.04 [18='eq_i686-linux 2.4.27-2-386'] 0.0000 0.1985 0.00 [20='eq_i686-linux 2.6.24.3'] 0.0000 0.1356 0.00 [21='eq_i686-linux-64int 2.6.22-1-k7'] 0.0000 0.1538 0.00 First of all, the const line has a Theta of zero, so we start with 0 and look for positive values. (It might come the opposite direction as well, with const being 1 and influencing factors having negative impact on the success). Where Theta is one (and T-test is >0) we may expect success. Where Theta is zero we must expect no success. Where Theta is in between we know that we had both success and failure on this operating system/version. Again, if you click on the link at the top of the regression table, you find the three tests for amd64-freebsd 6.2-release: 1094946 FAIL 715474 PASS 714951 PASS That gives 2/3 PASS => 0.0000+0.6667 expected value of the success. > I need a summary that tells me more clearly "this modules tends to > fail in these circumstances". I'll work this example into the about page later. Thanks for asking:) -- andreas
