[Numpy-discussion] is it a bug?

Vincent Schut schut@sarvision...
Fri Mar 13 05:13:27 CDT 2009


Travis E. Oliphant wrote:
> shuwj5460@163.com wrote:
snipsnip

Travis, thanks for the excellent explanation! It clears something which 
I think is related to this, I've been wanting to ask on the ml for some 
time already.
Now here's the case.
I often have 4d arrays that are actually related sets of satellite 
imagery, and have the form of [date, band, y, x]. These can get pretty 
large, so I like to prevent too much broadcasting or reshape-copy-ing 
when indexing to save some memory. However, I regularly have to apply 
some boolean index of [date, y, x] to each of the band dimensions (think 
of the bool mask as a threshold base on just one of the bands). 
Currently I usually loop over all band indices, e.g.

for b in data.shape[1]:
	data[:, b, :, :][mask] = 0

Would there be a way to do this in a more numpy-like fashion that is 
also memory-efficient?

Thanks,
Vincent.



More information about the Numpy-discussion mailing list