[Gretl-users] Re: Example code of 'eigensym' and way faster 'eiggen2'
On Sat, 11 Jan 2020, Sven Schreiber wrote:
Am 11.01.2020 um 21:26 schrieb Allin Cottrell:
> On Sat, 11 Jan 2020, Sven Schreiber wrote:
>>
>> So eigensym looks pretty bad "always". Since Ioannis had a newer PC,
>> this doesn't look like a pre-Haswell CPU issue. Perhaps something
>> Windows-specific (with OpenMP)?
>
> It's possible there was some specific issue with the openblas library
> we've been packaging for Windows, which was quite an old version. That's
> now replaced in today's snapshots with a new build of current OpenBLAS
> (version 0.3.7). I'd be interested to hear if that makes any difference.
The snapshot refuses to start, asking for libgfortran-5.dll or something.
Ah, OK. That's now included in the latest snapshots.
> Oh, one other point. At present eigensym() tests for symmetry of
the
> input. That's not going to affect the timing much, but maybe we should
> junk the test? In some other functions that expect symmetric input we
> don't bother to test -- it's up to the user.
Hm, don't know. I remember reading such a warning in the help for some
function. This warning would have to be copied to the eigensym doc I
guess. But for tiny matrices it may make a relative difference I guess.
But what would be the quickest or easiest user-testable way for checking
symmetry? (Can't test because my snapshot isn't starting ;-)
I think that shouldn't be a problem for the user: you should know if
a matrix is symmetric or not without having to test. Is it the
result of A'A, or a covariance matrix, or some such? In fact, what
actually matters for the relevant lapack functions is not that the
matrix is symmetric, but that it should be _taken as_ symmetric:
lapack will read only the lower or upper triangle. (I'll have to
check on this, but maybe we could standardize on, say, only the
lower triangle is ever read.)
Allin