Just so no one thinks I quit the project: I'm still working on the ARMA fitting routines. It's a major pain, and I haven't had a lot of time for it lately. I didn't check in any code because I want to finish an initial version before submitting it for review at savannah.
-Jason On Wed, Oct 18, 2006 at 08:00:21AM -0700, Ben Pfaff wrote: > I'm still working on getting data-in.c and data-out.c just right. > It's much more complicated to exactly reproduce SPSS output > formats than I thought. In particular, for F, COMMA, DOT, > DOLLAR, PCT, E, and the CC formats, I wrote a test that formats > about 350,000 different examples and ran it in both SPSS 13 and > PSPP. At first, about 100,000 of them differed. I've worked > diligently to make PSPP output more similar, and now I'm down to > about 55 that differ slightly. > > That doesn't include differences that fall into the following > categories: > > * 2,979 cases that differ slightly due to numerical > accuracy of formatting. SPSS isn't very good at this; > sometimes it will misformat relatively small integers > that are exactly represented as floating point numbers. > But other cases are not clear-cut. > > * 630 cases where SPSS will silently drop an initial 1 > that doesn't fit, e.g. 999.95 in F5.2 produces "000.0" > instead of " 1000". > > * 84 cases where SPSS just makes me say "WTF?" For > example, do SET CCB='-,[[[,]]],-' and then format 9.5 > as CCB8.0 and you get " *, 10 " instead of just > " 10 ". > > * Another 360 cases where SPSS makes me say "WTF?" for > another reason. For example, format -2 as E13.6 and you > get "*************" instead of "-2.00000E+000". > > * 360 cases where SPSS drops a negative sign. For > example, output -999.95 in F6.2 and SPSS produces > "1000.0" instead of " -1000". > > In other words, it's amazing how buggy SPSS 13 is. I wonder > whether these have been fixed in SPSS 15. > > Oh, and I've rewritten the system file reader. Much cleaner > now. Still needs careful testing. > > It'll probably be another week or two, at least, before I have > all of this cleaned up enough for review. > -- > Ben Pfaff > email: [EMAIL PROTECTED] > web: http://benpfaff.org > > > _______________________________________________ > pspp-dev mailing list > [email protected] > http://lists.gnu.org/mailman/listinfo/pspp-dev -- Jason H. Stover Assistant Professor Mathematics Department Georgia College & State University _______________________________________________ pspp-dev mailing list [email protected] http://lists.gnu.org/mailman/listinfo/pspp-dev
