Re: Random timing results with 'fsolve'

[Søren Hauberg]

ons, 12 10 2011 kl. 13:49 +0000, skrev bpabbott:
> On Oct 12, 2011, at 09:41 AM, Søren Hauberg <address@hidden> wrote:
> 
> 
> > Hi All
> > 
> > I am currently using 'fsolve' to solve a non-linear set of
> > equations,
> > for which it works great. However, when I try to determine the
> > running
> > time of the algorithm I get rather different results. I tried
> > 
> > for idx = 1:10
> > tic, [a1, b1] = fsolve (g, x, struct ("Jacobian", "off")); tid (idx)
> > = toc;
> > endfor
> > 
> > i.e. run the solver ten times with the same input and store running
> > times. After this I see
> > 
> > octave:313> max (tid_off) / min (tid_off)
> > ans = 4.2706
> > 
> > Does anybody know why I see so large differences in the timing
> > results?
> > 
> > Søren 
> > 
> 
> 
> Is the first iteration much slower than the rest?
> 
> 
> Perhaps parsing the code is much slower than actually minimizing the
> objective?

This does not appear to be the problem. Here are the timings for the
different iterations:

   27.259
   75.303
   82.480
   32.392
   19.313
   50.570
   42.955
   37.075
   67.188
   44.605

I don't really see any systematic behaviour here :-(

Søren