Explore using numba and cython

rigdenlab / SIMBAD

Sequence Independent Molecular replacement Based on Available Database

http://simbad.rtfd.io

BSD 3-Clause "New" or "Revised" License

3 stars 7 forks source link

Explore using numba and cython #116

Open hlasimpk opened 5 years ago

hlasimpk commented 5 years ago

This should improve speed of code when iterating through lists

fsimkovic commented 5 years ago

numba is not really an option unless CCP4 started supporting it. not sure where you'd need cython except for reading files though cctbx is about as good as it gets whilst keeping the Python interface

fsimkovic commented 5 years ago

I think you could gain some performance by

caching results. any calculation that is based on some input that might be identical to others could be stored in a cache (see functools.lru_cache). matthews/solvent/lattice scores/etc.

Apart from that you might struggle to keep the ability to distribute calculations both locally and across a cluster

hlasimpk commented 5 years ago

Fair point about numba. I thought I might get some performance improvements when looping through results, but this was just something I wanted to explore when I get a little more time rather than a hard plan

fsimkovic commented 5 years ago

had a stab in #117 to improve performance of the lattice search. I noticed that functools.lru_cache is not available until py3+ so you're stuck until CCP4 finally decide to upgrade their version.