[SciPy-User] Why optimize.minimize returns worse solution?

Pauli Virtanen pav@iki...
Wed Apr 10 12:53:16 CDT 2013


Jaakko Luttinen <jaakko.luttinen <at> aalto.fi> writes:
[clip]
> My gradient computations are incorrect at the moment, so that might be a
> reason for this problem. However, I still don't understand why the
> method returns a worse solution, even if the gradient computation is
> wrong. (And I'm not sure whether this is even the reason, or one of the
> reasons.)

There's probably a bug in the routines so that in the case
when a line search fails, they return a wrong value for the
objective function.

https://github.com/pv/scipy-work/commit/203334a5592

This affects only the case where the optimization fails ---
if the algorithm returns with success, then the problem
is not this one.

-- 
Pauli Virtanen



More information about the SciPy-User mailing list