Atomgroup index with None give back the full Atomgroup

[xiki-tempula]

Expected behavior

If I use None as atomgroup index, I expect to obtain an empty atom group.

>>> u = mda.Universe(crd)
>>> u.atoms[None]
<AtomGroup with 0 atoms>

Actual behavior

The original atomgroup is returned

>>> u = mda.Universe(crd)
>>> u.atoms[None]
<AtomGroup with 1 atom>
>>> u.atoms[None].names
array([['N', 'H1', 'H2', ..., 'CL', 'CL', 'CL']], dtype=object)

Code to reproduce the behavior

import MDAnalysis as mda
from MDAnalysis.tests.datafiles import PSF, DCD,  GRO, PDB, TPR, XTC, TRR,  PRMncdf, NCDF

u = mda.Universe(PSF, DCD)
u.atoms[None].names
#array([['N', 'HT1', 'HT2', ..., 'C', 'OT1', 'OT2']], dtype=object)

Current version of MDAnalysis

• Which version are you using? (run python -c "import MDAnalysis as mda; print(mda.__version__)") 2.0.0-dev0
• Which version of Python (python -V)? 3.8
• Which operating system? OSX

[IAlibay]

Mm this is indeed annoying behaviour, although as @richardjgowers points out, this is numpy default behaviour, i.e. passing None returns the entire array 😖

edit: to clarify, None returns what essentially amounts to newaxis so it's an array for the array, which is what's happening here.

[xiki-tempula]

@IAlibay well this doesn't explain

>>> u.atoms[None]
<AtomGroup with 1 atom>
>>> len(u.atoms[None])
1

If the entire array is returned.

[IAlibay]

Discussing this off-github with @richardjgowers, looks like the issue might be here?

mdanalysis/package/MDAnalysis/core/groups.py

Line 526 in 0de538e

return self._derived_class(self.ix[item], self.universe)

where the default Numpy behaviour is coming back to bite us (i.e. it returns an array of the array of ix, instead of the array of ix). There's probably a clean pythonic way to deal with this, but I can't think of one off the top of my head.

[jbarnoud]

In numpy, indexing with None is a shortcut for indexing with numpy.newaxis. This does not make sense for our groups. I think we should do the same as regular lists and raise a TypeError, even if it comes with an explicit test for None.

[zemanj]

In numpy, indexing with None is a shortcut for indexing with numpy.newaxis.

In fact, numpy.newaxis is None. So testing for None and raising an exception would definitely defeat using numpy.newaxis. Don't get me wrong, I don't see why using numpy.newaxis for indexing AtomGroups would be useful.
I'm all in favor of changing the behavior in one or the other way.

[IAlibay]

Sure TypeError sounds good to me :)

[jbarnoud]

In fact, numpy.newaxis is None.

Indeed. My point was that None creates a new axis which makes no sense for a group in mda. So I think we agree.