On 3/21/2011 1:03 PM, Robert Ramey wrote:
Andrew Sutton wrote:
If the community is interested in such a project, I have a minimalistic LP solver already written. I just need to "boostify" it and I'll upload within the next week.
I think you're underestimating the effort required to produce an LP library which would be considered acceptable to boost by two orders of magnitude.
In 3 moths? Absolutely! But I'm not terribly worried about that. If Chad good submits a proposal with reasonable goals, I'd be more than happy to fund the proposal.
Hmmm- two orders of magnitude = 100 so I was saying that that such a job would be 100 weeks. My real point is that to be accepted into boost, a library pretty much has to be demonstrably better than any opensource/openlicense alternative. I'm not making a judgment on this, it's just the way that I see the review process working.
Also, "toy implementations" (with few, if any, exceptions) have not passed the boost review process either. When something is submitted it is compared to all the alternatives and criticised according to the needs of a wide variety of applications so if it's not complete and very robust it doesn't get accepted.
Then there is a huge amount of work including build, test and documentation.
Finally, I think that numerical analysis issues such as near singular basis matrices, cycling, accumulated error and periodic re-inversion etc, etc are not issues that can be addressed satisfactorily in a couple of weeks for a library which would hope to reach the level that boost users expect and demand. I would say this is true even if the scope were confined to a simplex method implemenatation. Including mixed integer problems would be a whole 'nother level. To do this right really is a very big job.
I should note that many (most) of the GSOC proposals have failed to enter boost because the effort was way underestimated.
Of course, if anyone want's to take a shot, go ahead - it's a free country.
I agree with Robert. One of the best GSOC project, IMHO, is the Fusion project by Thomas Heller. As his mentor, I have to say that he has exceeded my expectation by 'ten orders of magnitude' :-) His work went beyond the GSOC time and until now and the future to come, he's totally devoted to the project and has become a full fledged author/maintainer. At the start, I had to make it clear that Thomas should expect more than GSOC's time frame and make him commit further beyond in order to fully satisfy the goals of the project --no less than what's expected of a Boost library in terms of quality. One year after GSOC 2010 and Thomas is still going strong. Bottom line: I think a key to a successful GSOC project is commitment by the student to go beyond the GSOC time frame. Regards, -- Joel de Guzman http://www.boostpro.com http://boost-spirit.com