Implement axis as tuple of int

[OGordon100]

Is your feature request related to a problem? Please describe.

Many numpy functions (e.g. mean) allow for axis= (or axes=) to be not just a single value, but a tuple of multiple axes.

Describe the solution you'd like

Allow functions such as detrend to accept multiple axis.

So instead of

data = np.random.rand(shape=(10,10))

data = scipy.signal.detrend(data, axis=0)
data = scipy.signal..detrend(data, axis=1)

we have

data = np.random.rand(shape=(10,10))

data = scipy.signal.detrend(data, axis=(0, 1))

Describe alternatives you've considered

Additional context (e.g. screenshots)

[Vijayhiga]

I am working on it.

[tupui]

I am working on it.

Thanks @Vijayhiga, but what do you mean? Are you looking specifically at scipy.signal.detrend or also other functions as suggested?

[rgommers]

I'm not sure this is desirable, it seems like a lot of busywork adding this for little real-world value for a function like detrend.

[OGordon100]

@rgommers completely disagree. Besides the fact that this could be carried over to other functions, there are lots of valuable use cases. If it's of no value, stuff it why was the functionality ever introduced anywhere else? It's also a really annoying inconsistency when half of functions can be called on multiple axes and half can't. Why not try and refine it?

In probe microscopy (encompassing a huge number of different techniques employed by pretty much every nanoscience group worldwide), for example, we have to correct our images for tilt as part of the most basic level of analysis. Groups are moving away from proprietary manufacturer software to python, and so I can tell you for a fact we have a number of codes calling several scipy functions twice, one for each axis. It would be useful to my sector of research, I can assure you 😉

[rgommers]

It's also a really annoying inconsistency when half of functions can be called on multiple axes and half can't.

It's way less than half I'd say.

for example, we have to correct our images for tilt

So what do you expect scipy.signal.detrend(data, axis=(0, 1)) to do? Just be a shorthand for:

data = scipy.signal.detrend(data, axis=0)
data = scipy.signal..detrend(data, axis=1)

or fit a 2-D plane and subtract that? I think if we implement it, we can only do the former because it needs to generalize to length-N input to the axis keyword. But better practice for your example would be to actually fit a plane.

signal functionality is basically 1-D, and making axis accept a tuple to do internally a repeated loop of 1-D operations isn't great. For proper n-D operations it's a different story.