Re: [eigen] Addition to Eigen-unsupported Spline module

[Hauke Heibel <hauke.heibel@xxxxxxxxx>]

Hi Jeff,

I don't remember anymore. I can only take care of this next week, or
maybe even the week after next week. I have two deadlines which I need
to take care of before I can handle the merge and other questions in
detail.

Thanks,
Hauke

On Fri, Jul 4, 2014 at 5:12 AM, Jeff J <complexzeros@xxxxxxxxx> wrote:
> Is there a reason for DerivativeType in SplineFwd.h to have a dynamic number
> of columns in the SplineTraits struct for fixed degree?
>
> typedef
> Array<_Scalar,_Dim,Dynamic,DerivativeMemoryLayout,_Dim,NumOfDerivativesAtCompileTime>
> DerivativeType;
>
>
>
>
> On Wed, Jun 25, 2014 at 7:06 PM, Jeff J <complexzeros@xxxxxxxxx> wrote:
>>
>> Thanks, I fixed the testing.
>>
>> Regarding PartialPivLU, I don't want to use that because, in general, the
>> matrix isn't guaranteed to be invertible, so I'm switching back to
>> HouseholderQR. It's good enough for my purposes.
>>
>>
>> On Wed, Jun 25, 2014 at 12:09 AM, Gael Guennebaud
>> <gael.guennebaud@xxxxxxxxx> wrote:
>>>
>>>
>>> Don't bother too much with the static assertion. We can add it later on,
>>> but basically you have to add an enum in Core/utils/StaticAssert.h and then
>>> use EIGEN_STATIC_ASSERT(NumTraits<T>::IsInteger, BLA_BLA_BLA).
>>>
>>> Regarding the accuracy, the unit frequently fails even with FullPivLU,
>>> and I did not observed any improvement of using the much more costly
>>> FullPivLU compard to PartialPivLU. So I'd rather move back to PartialPivLU
>>> and update the unit tests to:
>>>
>>> VERIFY_IS_APPROX(point,referencePoint);
>>> ...
>>> VERIFY_IS_APPROX(derivative, referenceDerivative);
>>>
>>>
>>> Moreover the "main" should iterate like this:
>>>
>>> void test_splines()
>>> {
>>>   for(int k=0; k<g_repeat; ++k)
>>>   {
>>>     CALL_SUBTEST( eval_spline3d() );
>>>     CALL_SUBTEST( eval_spline3d_onbrks() );
>>>     CALL_SUBTEST( eval_closed_spline2d() );
>>>     CALL_SUBTEST( check_global_interpolation2d() );
>>>     CALL_SUBTEST( check_global_interpolation_with_derivatives2d() );
>>>   }
>>> }
>>>
>>> cheers,
>>> gael
>>>
>>>
>>>
>>> On Tue, Jun 24, 2014 at 4:51 PM, Jeff J <complexzeros@xxxxxxxxx> wrote:
>>>>
>>>> Sorry, I'm not experienced with templates, but is there a preferred way
>>>> of doing such static assertion in Eigen? Maybe an example somewhere?
>>>>
>>>>
>>>> On Tue, Jun 24, 2014 at 5:01 AM, Hauke Heibel <hauke.heibel@xxxxxxxxx>
>>>> wrote:
>>>>>
>>>>> Ok, sounds good. For the moment, we can at least static assert that
>>>>> the Scalar of the array/vector is integral.
>>>>>
>>>>> Hauke
>>>>>
>>>>> On Tue, Jun 24, 2014 at 12:53 PM, Christoph Hertzberg
>>>>> <chtz@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
>>>>> > On 24.06.2014 12:24, Hauke Heibel wrote:
>>>>> >>
>>>>> >> What Christoph said backs up introducing a new typedef like e.g.
>>>>> >>
>>>>> >> EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(DenseIndex, Idx)
>>>>> >>
>>>>> >> Are there any objections about doing this? I am fine with it. That's
>>>>> >> all
>>>>> >> which keeps us from merging.
>>>>> >
>>>>> >
>>>>> > Well, I'm not objecting this, but I think it would be much easier to
>>>>> > simply
>>>>> > accept any type for derivativeIndices (the same as you do for
>>>>> > PointArrayType). Then the code would work for Eigen::ArrayXi as well
>>>>> > as
>>>>> > std::vector<short>, etc, and you would not need to introduce a new
>>>>> > typedef.
>>>>> > I doubt that anyone will have splines with more than 2^31 nodes in
>>>>> > the near
>>>>> > future, so requiring 64 bit indexes seems a bit over-engineered for
>>>>> > this
>>>>> > (for people requiring this it would be possible, nonetheless).
>>>>> >
>>>>> >
>>>>> > For a clean solution, I'd like to have some kind of static assertion
>>>>> > which
>>>>> > ensures `typename IndexArray` has a size() method, and an operator[]
>>>>> > which
>>>>> > returns an integer type. Later on specializations which determine if
>>>>> > the
>>>>> > size is fixed at compile-time might come handy.
>>>>> >
>>>>> >
>>>>> > Regards,
>>>>> >
>>>>> > Christoph
>>>>> >
>>>>> >
>>>>> >
>>>>> > --
>>>>> > ----------------------------------------------
>>>>> > Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
>>>>> > Cartesium 0.049
>>>>> > Universität Bremen
>>>>> > Enrique-Schmidt-Straße 5
>>>>> > 28359 Bremen
>>>>> >
>>>>> > Tel: +49 (421) 218-64252
>>>>> > ----------------------------------------------
>>>>> >
>>>>> >
>>>>>
>>>>>
>>>>
>>>
>>
>