[SciPy-user] Trouble with linsolve

[Nils Wagner]

Robert Cimrman wrote:
> Nils Wagner wrote:
>   
>> Robert Cimrman wrote:
>>     
>>> Nils Wagner wrote:
>>>   
>>>       
>>>> With UMFPACK 4.4 and a dense RHS I get
>>>>
>>>>     x = spsolve(K_dyn, f)
>>>>   File "/usr/lib64/python2.4/site-packages/scipy/linsolve/linsolve.py",
>>>> line 65, in spsolve
>>>>     return umf.linsolve( umfpack.UMFPACK_A, mat, b, autoTranspose = True )
>>>>   File
>>>> "/usr/lib64/python2.4/site-packages/scipy/linsolve/umfpack/umfpack.py",
>>>> line 566, in linsolve
>>>>     sol = self.solve( sys, mtx, rhs, autoTranspose )
>>>>   File
>>>> "/usr/lib64/python2.4/site-packages/scipy/linsolve/umfpack/umfpack.py",
>>>> line 508, in solve
>>>>     self._numeric, self.control, self.info )
>>>>   File
>>>> "/usr/lib64/python2.4/site-packages/scipy/linsolve/umfpack/_umfpack.py",
>>>> line 214, in umfpack_di_solve
>>>>     return __umfpack.umfpack_di_solve(*args)
>>>> ValueError: object too deep for desired array
>>>>     
>>>>         
>>>>>>> K_dyn
>>>>>>>           
>>>>>>>               
>>>> <71987x71987 sparse matrix of type '<type 'numpy.float64'>'
>>>>         with 3083884 stored elements (space for 3083884)
>>>>         in Compressed Sparse Column format>
>>>>
>>>> Any idea ?
>>>>     
>>>>         
>>> I use umfpack on a 32 bit system regularly without problems, so it may 
>>> be a 64-bit issue. I have yet to set up a 64-bit installation of scipy 
>>> to test it, though, so I cannot tell you more immediately.
>>>
>>> Did you also check your inputs to spsolve?
>>>
>>> r.
>>> _______________________________________________
>>> SciPy-user mailing list
>>> SciPy-user at scipy.org
>>> http://projects.scipy.org/mailman/listinfo/scipy-user
>>>   
>>>       
>> Robert,
>>
>> Just now I have filed two tickets. I hope someone can reproduce my results.
>>
>> http://projects.scipy.org/scipy/scipy/ticket/318
>> http://projects.scipy.org/scipy/scipy/ticket/319
>>
>> I have tried SuperLU on 32 and 64 bit machines - the segfault persists.
>> I have installed UMFPACK on a 64 bit machine.
>>
>> Any pointer would be appreciated.
>>
>> Thanks in advance.
>>     
>
> Nils, it was easy :) Both tickets can be resolved by:
>
> b = io.mmread('sherman2_rhs1.mtx').squeeze()
>
> in your test scripts. The rhs vector must be 1d array...
>
> I will update linsolve to do it automatically, but for now just squeeze 
> the extra dimensions.
>
> cheers,
> r.
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user
>
>   
Robert,

Thank you very much for your hint ! It works fine on 64 bit. Please can
you also add a docstring for spsolve.

Cheers,

Nils