[SciPy-dev] Trouble with optimize.fmin_ncg
dmitrey
openopt at ukr.net
Wed Jul 25 05:55:59 EDT 2007
Nils Wagner wrote:
> Matthieu Brucher wrote:
>
>> Hi,
>>
>> Here is a link to the diff :
>> http://projects.scipy.org/scipy/scipy/changeset?old_path=trunk%2FLib%2Foptimize&old=3116&new_path=trunk%2FLib%2Foptimize&new=3189
>> <http://projects.scipy.org/scipy/scipy/changeset?old_path=trunk%2FLib%2Foptimize&old=3116&new_path=trunk%2FLib%2Foptimize&new=3189>
>>
>> Matthieu
>>
>>
> Since the other optimizers are not affected, we may cut it down to
> optimize.py.
>
> Is it possible that
>
> http://projects.scipy.org/scipy/scipy/changeset/3176
> http://projects.scipy.org/scipy/scipy/changeset/3177
>
I have missed one bracket
fun(*(x,) +args) instead of fun(*((x,) +args))
(and now I fixed it, lines 1481, 1482,1027)
however, seems like it work without additional brackets as well:
>>> f = lambda x,y,z,t: x+y+z+t
>>> f(*((4,)+(5,)+(6,)+(7,)))
22
>>> f(*(4,)+(5,)+(6,)+(7,))
22
But since your funcs take no additional arguments (only x is present),
it's not the matter.
I checked - the problem remains.
I took a look at other changes and still can't find what's the problem.
the changes
[A,B] = numpy.dot([[dc**2, -db**2],[-dc**3,
db**3]],[fb-fa-C*db,fc-fa-C*dc])
309 d1 = empty((2,2))
310 d1[0,0] = dc**2
311 d1[0,1] = -db**2
312 d1[1,0] = -dc**3
313 d1[1,1] = db**3
314 [A,B] =
numpy.dot(d1,asarray([fb-fa-C*db,fc-fa-C*dc]).flatten())
are not related to the problem, because my debugger don't reach a
breakpoint placed here.
the changes
710 try:
711 rhok = 1 / (numpy.dot(yk,sk))
712 except ZeroDivisionError:
713 rhok = 1000.
715 try: # this was handled in numeric, let it remaines for
more safety
716 rhok = 1.0 / (numpy.dot(yk,sk))
717 except ZeroDivisionError:
718 rhok = 1000.0
719 print "Divide-by-zero encountered: rhok assumed large"
720 if isinf(rhok): # this is patch for numpy
721 rhok = 1000.0
are related to fmin_bfgs code.
> are responsible for the trouble with ncg ?
>
> Nils
>
> _______________________________________________
> Scipy-dev mailing list
> Scipy-dev at scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-dev
>
>
>
>
More information about the SciPy-Dev
mailing list