These are chat archives for thunder-project/thunder

23rd
Jul 2016
Jason Wittenbach
@jwittenbach
Jul 23 2016 02:03
@andins instead of using correlate try using cov
cov = data.toseries().flatten().cov().toarray()
plt.imshow(cov)
Be careful though: for 800x800 pixels, you’re looking at $(800^2)^2$ pairs of pixels. If the output comes in the form of float64 elements, then you’re looking at a correlation matrix that is easily > 3 TB in size. And this operation will be pulling that entire matrix back to your local machine for plotting.