can't change segmend ids in a loop for regions of a atomgroup

[kain88-de]

Expected behaviour

I want to iterate over a atomgroup and change the segids for regions in a loop.

Actual behaviour

Our getitem methods doesn't seem to honor the range I specify. So here when i=1
all atoms from 0-428 are getting the segid B.

Code to reproduce the behaviour

>>> import MDAnalysis as mda
>>> large_base_u = mda.fetch_mmtf('1bcc')
>>> n_copies = 5
>>> want_n_atoms = 214 * n_copies
>>> copy = large_base_u.select_atoms('name CA')[:want_n_atoms]
>>> letters = 'ABCDEFGH'
>>> for i in range(n_copies):
>>>     start = i*(u.atoms.n_atoms)
>>>     stop = (i+1)*u.atoms.n_atoms
>>>     print(letters[i], start, stop)
>>>     copy[start:stop].segments.segids = letters[i]
>>>     print(copy[:stop].segids)

>>> np.unique(copy.segids)
array(['C', 'D', 'E'], dtype=object)

Currently version of MDAnalysis:

(run python -c "import MDAnalysis as mda; print(mda.__version__)")
dev version

[kain88-de]

copy[0:214].segments.segids = 'A'
copy[214:428].segments.segids = 'B'

This is shorter. Here the second line is overwriting the atoms from 0-214

Any idea of why this happens. This makes it pretty hard to build my system currently

[kain88-de]

affects 0.16.2 as well

[richardjgowers]

@kain88-de setting segids won't create new Segments (any more). So I think what's happening is that u.atoms[:214].segments refers to the same segments as u.atoms[214:428].segments, so you're just setting segids to the same segments twice.

If you want to add a Segment to the Universe, do new = u.add_Segment(segid='B'); copy[214:428].segments = new

[kain88-de]

But wouldn't it be better then if segids are readonly? The simple fact that u.atoms[:214].segments.segids = 'a' doesn't raise an error lead me to believe it's a valid way to update segment ids.

Also you explanation makes sense. It's a bit confusing though that it works only half now. If I change all segids in a atomgroup to the same letter everything seems fine.

[richardjgowers]

@kain88-de it is a valid way to update segids. You're only running into trouble because you're updating the segid of the same segment twice. See the toy example below where I deliberately create two ags (ag1 and ag2) which both belong to the same segment (seg)

There's no real way to disallow setting the segid from an AtomGroup like this, because u.atoms.segments is just a SegmentGroup with no memory/knowledge that it came from an AtomGroup.

In [1]: import MDAnalysis as mda

In [2]: u = mda.Universe('adk_oplsaa.gro')
/home/richard/miniconda2/envs/mdadev/lib/python2.7/site-packages/MDAnalysis/topology/guessers.py:56: UserWarning: Failed to guess the mass for the following atom types: M
  "".format(', '.join(misses)))

In [3]: u.segments
Out[3]: <SegmentGroup with 1 segment>

In [4]: ag1 = u.segments.atoms[:100]

In [5]: ag2 = u.segments.atoms[100:]

In [6]: seg = u.segments[0]

In [7]: seg.segid
Out[7]: 'SYSTEM'

In [8]: ag1.segments.segids = 'A'

In [9]: ag1.segids
Out[9]: 
array(['A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A',
       'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A', 'A'], dtype=object)

In [10]: seg.segid
Out[10]: 'A'

In [11]: ag2.segments.segids = 'B'

In [12]: ag2.segments.segids
Out[12]: array(['B'], dtype=object)

In [13]: ag2.segids
Out[13]: array(['B', 'B', 'B', ..., 'B', 'B', 'B'], dtype=object)

In [14]: seg.segid
Out[14]: 'B'

In [15]: ag1.segids
Out[15]: 
array(['B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B',
       'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B', 'B'], dtype=object)

[orbeckst]

For docs we need a section on topology manipulation; relabeling is a common theme.