Re: [Crm114-general] -11 errors

[Robert K. <cr...@al...>]

 >
 >   From: Robert Klep <cr...@al...>
 >
 >    >This one does call out individual file statuses and pR's, so the
 >    >format of the :stats: variable has changed.
 >
 >   Just for my understanding: I'm doing a 'shortcircuited' classify() (F1
 >   F2 | F3 F4), and this is the result:
 >
 >   CLASSIFY fails; success probability: 0.5000
 >   Best match to file #0  prob: 0.2500  pR: -0.4771
 >   file #0: features: -1610612735, hits: 0, prob: 2.5000e-01, pR: -0.4771
 >   file #1: features: -1610612735, hits: 0, prob: 2.5000e-01, pR: -0.4771
 >   file #2: features: 536870913, hits: 0, prob: 2.5000e-01, pR: -0.4771
 >   file #3: features: 268435457, hits: 0, prob: 2.5000e-01, pR: -0.4771
 >
 >   Am I correct in stating that file #2 (that would be F3) is the winner?
 >
 >No, the one you should take is the one under "best match", second line,
 >the one that looks like:
 >
 >   Best match to file #0  prob: 0.2500  pR: -0.4771
 >
 >But you've found a test case that's REALLY broken.

Since I was getting #0 as winner all the time, even after shuffling the
order of the files, I figured as much.

 >Files F1 and F2 both show _negative_ numbers of hits, and that's way
 >impossible unless there's a bug.
 >
 >How did you get this?

Well, I realised I'm using .css files which were made on x86, but the
results are created by using them on PPC (which has a different endianess).

If that might sound like a plausible reason, I'll take all the blame and go
sit in a corner for a while ;)

                                                                     robert