[SciPy-Dev] Powell failure on MingW windows build - any insights?

[Matthew Brett]

On Wed, Apr 23, 2014 at 3:52 PM, Matthew Brett <matthew.brett at gmail.com> wrote:
> Hi,
>
> On Wed, Apr 23, 2014 at 11:29 AM, Pauli Virtanen <pav at iki.fi> wrote:
>> 23.04.2014 02:17, Matthew Brett kirjoitti:
>> [clip]
>>>> Having the check that far (34 evals) in the tail may however be
>>>> excessive. The other routines check earlier only after a few
>>>> iterations, so if you can see how much of the trace is
>>>> reproducible, that should be OK.
>>>
>>> The course is the same through iteration 14 - is that acceptable as
>>> a length?
>>
>> Should be OK.
>
> Here are values 0:15
>
> array([[ 0.        ,  0.        ,  0.        ],
>        [ 0.        ,  0.        ,  0.        ],
>        [ 1.        ,  0.        ,  0.        ],
>        [-1.618034  ,  0.        ,  0.        ],
>        [ 0.        ,  0.        ,  0.        ],
>        [-0.61803397,  0.        ,  0.        ],
>        [ 0.381966  ,  0.        ,  0.        ],
>        [ 0.190983  ,  0.        ,  0.        ],
>        [ 0.61803397,  0.        ,  0.        ],
>        [ 0.76393201,  0.        ,  0.        ],
>        [ 0.85410196,  0.        ,  0.        ],
>        [ 0.70820392,  0.        ,  0.        ],
>        [ 0.79837386,  0.        ,  0.        ],
>        [ 0.74264577,  0.        ,  0.        ],
>        [ 0.72949016,  0.        ,  0.        ]])
>
> I wonder if it matters that this only covers the search for the first
> value in the vector?

I guess an alternative would be to have several possible paths, say up
until 30, where we have agreed that differences between these paths
are due to acceptable differences in rounding error.   Was there a
reason to choose 34:39 previously?  For example, was this a constant
part of the path across machines, compared to other parts?

Cheers,

Matthew