On Tue, Mar 29, 2011 at 11:35 PM, Matthew Brett <[email protected]> wrote:
> Hi,
>
> On Tue, Mar 29, 2011 at 11:43 AM, Andy Ray Terrel <[email protected]> 
> wrote:
>> On Tue, Mar 29, 2011 at 7:22 PM, Aaron S. Meurer <[email protected]> wrote:
>>>
>>> On Mar 29, 2011, at 12:51 AM, Andy Ray Terrel wrote:
>>>
>>>> On Tue, Mar 29, 2011 at 9:13 AM, Tim Lahey <[email protected]> wrote:
>>>>> On 03-29-2011, at 1:44 AM, Andy Ray Terrel wrote:
>>>>>
>>>>>> There are some major confusions on this list about the difference
>>>>>> between algorithms and storage.
>>>>>
>>>>> I'm not confused and I'm not sure if other people are. I just think 
>>>>> storage alone isn't suitable for a GSoC project. Certainly one could just 
>>>>> implement storage techniques behind the scenes and no algorithms, but 
>>>>> that would only save space, not computation time.
>>>>
>>>> On modern hardware, computation is free memory is expensive.
>>>>
>>>>>
>>>>>> Yes, it is typical to use iterative
>>>>>> algorithms with sparse matrices and factorizations with dense, but
>>>>>> this is because sparse matrices are typically very large (1M -- 10^12
>>>>>> entries with 1000 -- 2B non-zeros).  One will switch to things that
>>>>>> don't fill in the matrix but SymPy will probably never be at this
>>>>>> level.
>>>>>
>>>>> With symbolic matrices, the computation time for each calculation is much 
>>>>> higher than with numerical matrices so the benefit of sparse matrix 
>>>>> techniques occur at much smaller matrix sizes.
>>>>>
>>>>>>  One can implement some Krylov subspace iteration algorithms,
>>>>>> but its somewhat pointless (how do you evaluate a residual in
>>>>>> symbols).  Another algorithm ILU versus LU is also silly (basically it
>>>>>> throws away small fill in entries but we have no way of saying what is
>>>>>> small in symbolics).  For this reason I suggest just implementing
>>>>>> standard factorizations and dealing with the fill in as it happens if
>>>>>> we get to the point that we need to do UMFPACK style direct
>>>>>> factorizations we can do that the next summer.  We practitioners only
>>>>>> use "algorithms suited for sparse matrices" because of the size of the
>>>>>> matrices which will not be an issue in SymPy.
>>>>>
>>>>> I've already stated that storage isn't suitable for a GSoC project.
>>>>
>>>> I respectfully disagree.  Furthermore I find it distasteful for one
>>>> applicant to tell another that their idea is not suitable.
>>>
>>> Let's remain civil.  I think it is important to decide what is needed and 
>>> what is less needed, or what will be better than the current (dense) 
>>> implementation and what will actually end up being worse.
>>
>> Please give me your guide of civility so I can rip out every page and
>> use it as toilet paper.
>
> I don't think that will serve you well, civility is very important:
>
> http://www.codesimplicity.com/post/open-source-community-simplified/ -
> see 'Encourage a total absence of personal negativity'.
> http://producingoss.com/en/setting-tone.html see 'Nipping rudeness in the bud'
>
> Best,
>
> Matthew
>

Matthew, I would agree that civility is important, but any civility
guide that tells a person not to speak up with another  is attacking a
student's proposal is as useful as toilet paper.


This thread has become a diatribe of opinions.  Nevertheless, Frank,
you should submit this proposal and continue to submit a patch for the
GSOC application.

-- Andy

-- 
You received this message because you are subscribed to the Google Groups 
"sympy" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/sympy?hl=en.

Reply via email to