On Thu, Nov 12, 2009 at 3:30 PM, Charles R Harris <charlesr.har...@gmail.com > wrote:
> > > On Thu, Nov 12, 2009 at 12:52 PM, Christopher Barker < > chris.bar...@noaa.gov> wrote: > >> Robert Kern wrote: >> >> > Didn't we already do this? >> > >> > http://www.mail-archive.com/numpy-discussion@scipy.org/msg21010.html >> >> Indeed we did. What I posted then ( and have improved a bit now). Is a >> Python version. Written in Python, it has an advantage of using less >> memory for a big array, but is slower in other respects than a Python >> list. This is probably why we all use lists for this when we need it! >> >> > IIRC, python lists *are* expanding arrays of contiguous memory, not linked > lists. They are efficient for appending and indexing, less so for deletion > and insertion. It's one of those tradeoffs. So what you buy with an array > implementation is the space/time efficiency of not having to allocate python > types to put on the list. But you probably need to go through a python type > at some point anyway, so... > > And here is a pep <http://www.python.org/dev/peps/pep-3128/%20> for a Blist with a table of some of the performance tradeoffs. I think you could use small ndarrays in the leaf nodes and get around having to handle all the type stuff. Chuck
_______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion