Parallel computing tutorial¶

try pytraj online:

Using multiprocessing¶

Task

Perform RMSD calculation in parallel

In [1]:

import warnings
warnings.filterwarnings('ignore', category=DeprecationWarning)
import pytraj as pt

traj = pt.iterload('tz2.nc', 'tz2.parm7')
print(traj)

pytraj.TrajectoryIterator, 101 frames: 
Size: 0.000503 (GB)
<Topology: 223 atoms, 13 residues, 1 mols, non-PBC>

In [2]:

pt.pmap(pt.rmsd, traj, mask='@CA', ref=traj[0], n_cores=4)

Out[2]:

OrderedDict([('RMSD_00001',
              array([  1.94667955e-07,   2.54596866e+00,   4.22333034e+00, ...,
                       4.97189564e+00,   5.53947712e+00,   4.83201237e+00]))])

Using MPI¶

Task

Perform RMSD calculation in parallel

In [3]:

%%file my_script.py

## create a file name my_scrip.py (you can use any name you like)

import pytraj as pt
from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.rank

# load files
traj = pt.iterload('tz2.nc', 'tz2.parm7')

# call pmap_mpi for MPI

# we dont need to specify n_cores=6 here since we will use `mpirun -n 6`
data = pt.pmap_mpi(pt.rmsd, traj, mask='@CA', ref=traj[0])

# data is sent to rank==0
if rank == 0:
    print(data)

# run

Overwriting my_script.py

In [4]:

# run in shell

! mpirun -n 4 python my_script.py

OrderedDict([('RMSD_00001', array([  1.94667955e-07,   2.54596866e+00,   4.22333034e+00, ...,
         4.97189564e+00,   5.53947712e+00,   4.83201237e+00]))])

(tutorial_parallel.ipynb; tutorial_parallel_evaluated.ipynb; tutorial_parallel.py)