On Wed, 7 Nov 2012, Pindar wrote: > when working on my multivariate statistic package I found the following bug > in function msortby: > If one wants to sort an already sorted matrix the matrix elements get though > rearranged. > In the 2-factor Anova with interaction this results in wrong calculations. > > <hansl> > nulldata 24 > #two factor orthogonal Anova > matrix FaktorA={1;1;1;1;1;1;2;2;2;2;2;2;3;3;3;3;3;3;4;4;4;4;4;4} > matrix FaktorB={1;1;2;2;3;3;1;1;2;2;3;3;1;1;2;2;3;3;1;1;2;2;3;3} > matrix > OutputM={69;77;61;67;65;69;79;83;62;64;72;78;80;78;74;76;80;74;75;67;70;62;70;76} > > eval OutputM~FaktorA > eval msortby(OutputM~FaktorA,2) > <hansl> Hmm, I'm not seeing anything wrong with gretl's output on this. I take it your point is that, since FaktorA is already in sort order, msortby() should be a null operation and the two "evals" above should give the same output -- but that in fact they don't? Here they do give the same. To check explicitly: <hansl> matrix FaktorA={1;1;1;1;1;1;2;2;2;2;2;2;3;3;3;3;3;3;4;4;4;4;4;4} matrix FaktorB={1;1;2;2;3;3;1;1;2;2;3;3;1;1;2;2;3;3;1;1;2;2;3;3} matrix \ OutputM={69;77;61;67;65;69;79;83;62;64;72;78;80;78;74;76; \ 80;74;75;67;70;62;70;76} matrix K = OutputM ~ FaktorA matrix P = msortby(K, 2) Test = K - P print Test </hansl> Here the Test matrix is all zeros. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Wed, 7 Nov 2012, Pindar wrote: > Indeed, on my W7 PC with the new snapshot the 'evals' are > not the same. The order within the bins is different. I > expected the null operation and checked other parts of the > code until I found the reason explained. I gonna try with > 'matrix K' instead of the ~ operation and report then. OK, I suspect this must be a difference between the implementation of the qsort() function in the Microsoft C library and the GNU C library. The C standard doesn't specify whether or not elements of an array passed to qsort() that compare equal will remain in the original order. Apparently glibc is preserving the original order while MS is messing with it.

In many contexts I guess this won't make any relevant difference, but in some contexts it can. If it's important to avoid reordering of elements that compare equal we can use a fancier comparison function that looks at the addresses of the elements when the values compare equal. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users