New submission from Mike <mikha.my...@gmail.com>:
I have a random bug with scipy's optimize .
I use it in the context of SageMath (Python 3.7.3).
I checked 3 algorithms : shgo, dual_annealing and full_optimize.
All don't work well (at all !).
I optimise with a 3 parameters functions with given bounds. The algorithm
randomly sends to my error function the "1e-8" value for any of the 3
parameters. I displayed the vector + error to understand it. AS you can see, in
this case, the first parameter is 1e-8 whereas it is outside the bounds.
Sometimes it happens for the second parameter, or the 3rd ....
------
[3.95049282e-19 3.03055607e-20 1.96212098e+29] 0.0030238733573031864
[3.95273920e-19 3.05821352e-20 1.90997635e+29] 0.002957956545311753
[3.95037412e-19 3.04080173e-20 1.93312145e+29] 0.0029572689364709224
sage.all_cmdline:33: IntegrationWarning: The occurrence of roundoff error is
detected, which prevents
the requested tolerance from being achieved. The error may be
underestimated.
[1.00000000e-08 3.04080173e-20 1.93312145e+29] 1.0713560755245306
Overflow
[3.95037412e-19 3.04080173e-20 1.93312145e+29] 0.0029572689364709224
-----
The behaviour is the same for all of them. However the value inserted is not
always the same, I noticed 3 : 1e-10, 1e-8 and 1.49011612e-08
Moreover, all do not send nice values. For example, the full_optimize always
send values close to the highest bound. Perhaps are the value computed with a
simple linear average and I guess it should not be the case.
my bounds are :
bounds = [(1*e,4*e),(0.1*e,0.5*e),(1e5,1e40)]
with e=1.6e-19
Is there an alternative for optimization in python ?
regards,
Mike
----------
messages: 368540
nosy: mmyara
priority: normal
severity: normal
status: open
title: scipy's differential_evolution bug
versions: Python 3.7
_______________________________________
Python tracker <rep...@bugs.python.org>
<https://bugs.python.org/issue40581>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
