Mass COM calculation

[richardjgowers]

So over in #670 we mooted having a mass COM method, so I mocked this up quickly in Cython. (it's roughly 30x faster than previous)

Basically, given a list of 0 based resids, returns the center of geometry for each resid. So is the equivalent of:
np.array([r.center_of_geometry() for r in u.residues])

So there's lots of rough edges as to how the function works, including adding a mass (and charge?) weighting option so it returns CoM.

So my question is, is this as good as cython as we can write? (maybe @kain88-de or anyone better at cython than me can say)

I did a numpy typed version:
1206092

And a memory view version:
62fdf99

And couldn't see any difference in the performance

[kain88-de]

You can tell the compiler how the array is stored (Row or Column Major). That can be done by giving it the strides directly float[:, ::1 is row major for example the C/Python standard. That should give a small boost but requires that the given arrays follow that memory layout.

[kain88-de]

I would keep your current version as this is less pain for someone who uses the function.

[kain88-de]

What was your benchmark? I don't see anything greatly wrong here so it should be fine and generally faster then a python version.

Fix the doc string and it should be good to go.

[hainm]

you can turn this off globally in setup.py file. check my pytraj's one

https://github.com/Amber-MD/pytraj/blob/master/setup.py#L327-L347

[richardjgowers]

@kain88-de I'm loading the PSF/DCD system and doing

In [1]: import MDAnalysis as mda

In [2]: u = mda.Universe('adk.psf','adk_dims.dcd')

In [3]: from MDAnalysis.lib.trans import residue_positions

In [4]: indices = u.atoms.resids - 1

In [5]: %timeit out = residue_positions(u.atoms.positions, indices)
10000 loops, best of 3: 140 µs per loop

In [6]: %timeit long_way = np.array([r.center_of_geometry() for r in u.residues])
100 loops, best of 3: 3.03 ms per loop

I'll look at defining the layout and see..

[kain88-de]

You don't need to define the layout. That was more as an FYI. It should be absolutely fine here as it is.