[Numpy-discussion] A numpy accumulator...
Tue Oct 6 00:15:31 CDT 2009
Tom K. wrote:
> Chris - I really like this and find it useful. I would change the name to
> something like "growable" or "ArrayList"
hmm. I think I like "growable" or maybe "growarray".
> I think the right amount to grow is 2x -
I think that may be too much.. one if the key advantages of this over
python lists is that there should be a memory use advantage -- when you
are pushing memory bounds, using twice what you need is a bit much.
> Perhaps the configurable grow ratio is a good
> thing, although giving a knob means people are going to set it wrong.
maybe, but most folk will use the default anyway. I'm certainly going to
keep it configurable while under development -- the better to benchmark
> I would also vote "+1" for an ND version of this (growing only a single
Yes, I think that is a good idea, and would certainly be useful for a
common case -- growing a table of data, perhaps when reading a file, etc.
> Keeping 2x for each of n dimensions, while conceivable, would
> be 2**n extra memory, and hence probably too costly.
That, and the fact that you'd have to move a bunch of memory around as
it grew -- if you only grow the first dimension (for C order, anyway),
you can just tack stuff on the end (which usually necessitates a copy
anyway, but it still seems easier.
thanks for the feedback,
Christopher Barker, Ph.D.
Emergency Response Division
NOAA/NOS/OR&R (206) 526-6959 voice
7600 Sand Point Way NE (206) 526-6329 fax
Seattle, WA 98115 (206) 526-6317 main reception
More information about the NumPy-Discussion