Re: [Gretl-users] question - Unstable regression results
On Thu, 28 Apr 2016, Sven Schreiber wrote:
Am 28.04.2016 um 17:54 schrieb Allin Cottrell:
> It's more surprising that you could get differences on the same machine
> (with the same OS and the same pre-built gretl version), on different
> days. However, I've read that on MS Windows certain software, such as
> DirectX, may reset the precision of floating-point units. So it may be
> that -- when the results are so terribly sensitive to precision -- what
> gets printed by gretl depends on what else you've been running lately.
Interesting! Who would have thought...
You can find a lot about this in the blogosphere, a good deal of it
from (frustrated) writers of games who want to be able to offer exact
"replays".
Footnote: when we see apparently indeterministic results, the first
suspicion is that the code may somehow be reading from uninitialized
memory. So that was the first thing I checked, and valgrind told me it
was not the case here.
Another footnote: I've tried Mikael's regression on two different
machines running 64-bit Linux over two days, and I haven't yet seen
anything but 2.68356 for the "troublesome" coefficient.
Allin