MAINT: speed up 1-norm estimation for non-tiny linear operators

[argriffing]

This also halves the expm_multiply time in $ python runtests.py --bench -s sparse.linalg on my system.

[larsmans]

LGTM.

[argriffing]

@ewmoore just to be clear, are you suggesting to change

g = np.sort(g)[::-1]

to

g.sort()
g = g[::-1]

with the idea that it avoids a copy?

[ewmoore]

Yes, exactly. Calling the method performs an in place sort. Calling the
function returns a sorted copy. Reversing the sorted array results in a
view either way so is always cheap.

On Tuesday, October 28, 2014, argriffing notifications@github.com wrote:

@ewmoore https://github.com/ewmoore just to be clear, are you
suggesting to change

g = np.sort(g)[::-1]

to

g.sort()
g = g[::-1]

with the idea that it avoids a copy?

—
Reply to this email directly or view it on GitHub
#4095 (comment).

[argriffing]

@ewmoore thanks, I've improved the sorting following your suggestion

[argriffing]

TravisCI errors are weird Cython errors, maybe related to a recent Cython release and its interaction with Python3?

[ev-br]

The weather in the cloud seems better now: restarted build is all green.

[argriffing]

Closing this, to break it into parts so that it is easier to review and merge.