Re: Study about accuracy of statistical software, incl. Octave

[Marco Caliari]

Dear Jaroslav,

I read the interesting paper http://www.imstat.org/bjps/papers/Almiron.pdf
The attached patch (against 3.1.54) fixes the less important deficiency in 
the computation of the mean: it divides the vector by its maximum value before computing 
the mean. With the patch, the example NumAcc4 in Table 1 of the paper has 
15 correct digits. Moreover, now it is possible to compute

mean([1.7e308,1.6e308])

getting the right answer and not Inf. The patch also works for matrices, 
but I don't know if the implementation is optimal.
The other deficiencies are much harder to fix. I will give a look.

Best regards,

Marco

mean.diff
Description: Text Data