Re: r25228 - in /trunk/test_suite/shared_data/curve_fitting/profiling: profiling_relax_fit.py relax_fit.py -- August 25, 2014 - 11:07

Hi Edward.

This post continues from:
http://thread.gmane.org/gmane.science.nmr.relax.devel/6814

As I wrote in this post, the new user function:
"relax_disp.r2eff_estimate()"

gives freedom to the user to try other algorithms.
Here scipy.optimize.leastsq which should be (or is) a well known
option in the python community for minimisation.

Did you find errors in MINPACK or in scipy ?

Best
Troels


2014-08-25 10:13 GMT+02:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
> Hi Troels,
>
> You should note that the 20x speed up is due to the change in
> optimisation algorithm rather than scipy vs. minfx vs. MINPACK.  The
> reason I wrote minfx is that I started with scipy optimisation but
> found all implemented algorithms contained fatal bugs.  This was not
> fixed for years after I reported it, and I don't know if the code had
> changed since 2003 when I looked into it.  Anyway a scipy optimisation
> solution is incompatible with the minfx optimisation solution in
> relax.  If you derive and code the gradients into relax, then you can
> use the minfx LM algorithm as a solution.
>
> Regards,
>
> Edward
>
>
>
>
>
>
>
> On 25 August 2014 01:08,  <tlinnet@xxxxxxxxxxxxx> wrote:
> > Author: tlinnet
> > Date: Mon Aug 25 01:08:44 2014
> > New Revision: 25228
> >
> > URL: http://svn.gna.org/viewcvs/relax?rev=25228&view=rev
> > Log:
> > Further improved the profiling of relax curve fit.
> >
> > This profiling shows, that Python code is about twice as slow as the C 
> > code implemented.
> >
> > But it also shows that optimising with scipy.optimize.leastsq is 20 X 
> > faster.
> > It also gives reasonable error values.
> >
> > Combining a function for a linear fit to guess the initial values, together
> > with scipy optimise, will be an extreme time win for estimating R2eff 
> > values fast.
> >
> > A further test would be to use relax Monte-Carlo simulations for say 
> > 1000-2000 iterations,
> > and compare to the errors extracted from estimated covariance.
> >
> > Added:
> >     trunk/test_suite/shared_data/curve_fitting/profiling/relax_fit.py
> > Modified:
> >     
> > trunk/test_suite/shared_data/curve_fitting/profiling/profiling_relax_fit.py
> >
> > [This mail would be too long, it was shortened to contain the URLs only.]
> >
> > Modified: 
> > trunk/test_suite/shared_data/curve_fitting/profiling/profiling_relax_fit.py
> > URL: 
> > http://svn.gna.org/viewcvs/relax/trunk/test_suite/shared_data/curve_fitting/profiling/profiling_relax_fit.py?rev=25228&r1=25227&r2=25228&view=diff
> >
> > Added: trunk/test_suite/shared_data/curve_fitting/profiling/relax_fit.py
> > URL: 
> > http://svn.gna.org/viewcvs/relax/trunk/test_suite/shared_data/curve_fitting/profiling/relax_fit.py?rev=25228&view=auto
> >
> >
> > _______________________________________________
> > relax (http://www.nmr-relax.com)
> >
> > This is the relax-commits mailing list
> > relax-commits@xxxxxxx
> >
> > To unsubscribe from this list, get a password
> > reminder, or change your subscription options,
> > visit the list information page at
> > https://mail.gna.org/listinfo/relax-commits
>
> _______________________________________________
> relax (http://www.nmr-relax.com)
>
> This is the relax-devel mailing list
> relax-devel@xxxxxxx
>
> To unsubscribe from this list, get a password
> reminder, or change your subscription options,
> visit the list information page at
> https://mail.gna.org/listinfo/relax-devel