On 12/13/2009 08:32 AM, anand jeyahar wrote: > A crucial element in this hypothetical module would be the main graph > data structure. The simplest approach would be to implement it in pure > python, with lists, dicts and such, as many libraries do. However, this > would rule out its use by high-performance code, which would need a > simpler C-based data structure for direct interaction. On the other > hand, I'm not sure if there is a need for a high performance graph > module in python's standard library... > > I disagree...I am not sure of the current need in terms of a precise > survey.....But IMO, as bearophile pointed out.....networkx is the most > popular........and from their claims it is targeted at mathematicians, > physicists, biologists, computer scientists, social scientists. > > Given the current trend in the growth of social and professional > networks..... and social scientists (Disclaimer: i aspire to be one).. i > do expect a growing demand for graph data structures and high > performance ones soon enough.. > > so IMHO if we are going in for it we should go for the high performance > graphs too..
I certainly think it is important to have a high-performance graph library, that is why I started to write one. I was talking simply about its inclusion in the standard library. Since something as basic for scientific computing as, for instance, numpy does not belong in the standard library, I don't think it makes sense to push for the inclusion of a scientific graph library. Could we (or rather should we) even make it _independent_ of numpy? But if the idea is to consolidate the existing graph libraries into an uniform package, I'm certainly not against it in principle. But I think we should first familiarize ourselves with the existing variants, to see if there is really a common ground. The basic interface to the graph data structure should not vary much, I believe, but the implementation may. For instance, I chose in graph-tool to implement most of it in C++ with the Boost Graph Library... When I get some time, I'll take a careful look at the implementations listed in this thread, some of which I don't yet know. Cheers, Tiago
signature.asc
Description: OpenPGP digital signature
-- http://mail.python.org/mailman/listinfo/python-list
