Conversation
|
@hmacdope should we emphasize more that people should try and install to make use of the much faster distance calculations? Or should we include it as a conda package run time dependency for x86 architectures? |
|
Distopia works on my old x86 macOS: In [1]: import MDAnalysis as mda; import MDAnalysisTests.datafiles as data
In [2]: u = mda.Universe(data.TPR, data.TRR)
In [3]: u.atoms.n_atoms
Out[3]: 47681
In [10]: %timeit d = mda.lib.distances.distance_array(u.atoms[:5000], u.atoms[500:4000],
⋮ box=u.dimensions)
629 ms ± 2.78 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
In [11]: %timeit d = mda.lib.distances.distance_array(u.atoms[:5000], u.atoms[500:4000],
⋮ box=u.dimensions, backend="distopia")
324 ms ± 163 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)and gives 2x speed-up for a medium-sized problem. For all 47k atoms ... In [14]: %timeit d = mda.lib.distances.distance_array(u.atoms, u.atoms, box=u.dimensions, backend="distopia")
Segmentation fault: 11oops.... but in all fairness, the serial version ran my laptop out of its 16 GB of memory but completed In [4]: %timeit d = mda.lib.distances.distance_array(u.atoms, u.atoms, box=u.dimensions,
⋮ backend="serial")
1min 36s ± 3.02 s per loop (mean ± std. dev. of 7 runs, 1 loop each) |
|
The post could now be published in principle (edit: because we now have 3.13 cf packages) but let’s wait until at least tomorrow in case anyone wants to make any changes. |
Co-authored-by: Irfan Alibay <IAlibay@users.noreply.github.com>
(MDAnalysis/mdanalysis#4805) (or it will have them soon enough so that I can remove the limitation)
Co-authored-by: Jenna M Swarthout Goddard <38287809+jennaswa@users.noreply.github.com>
b14be95 to
3f24e89
Compare
|
|
||
| * Support for Gromacs v2024.4 and v2025 TPR files. | ||
| * A new "water" keyword for selecting water residues. | ||
| * Support for **[distopia][] 0.4.0** as an optional distance library |
There was a problem hiding this comment.
@hmacdope can you look over the distopia part, please?
There was a problem hiding this comment.
see also PR MDAnalysis/mdanalysis#4970 for doc update
- installation instructions - link to docs for backends - mini example - mention (currently undocumented) MDAnalysis.lib.distances.HAS_DISTOPIA
463f1ef to
3199858
Compare
|
@jennaswa with release post up, are you able to put out announcements (LinkedIn, bsky, discussions)? (Not sure if I have any time soon) |
No description provided.