Saving frames in parallel

import xarray as xr
from xmovie import Movie

ds = xr.tutorial.open_dataset('air_temperature').isel(time=slice(0,200))

# Creating the movie object
mov = Movie(ds.air, vmin=230, vmax=310)

The creation of a movie can take quite long for datasets with many timesteps, creating many frames in a loop.

%%time
mov.save('movie.mov', overwrite_existing=True)

Movie created at movie.mov
CPU times: user 34.7 s, sys: 1.02 s, total: 35.7 s
Wall time: 59.4 s

You can speed up the frame creation by activating the parallel option. This will save the frames using dask.

For this to work you need to chunk the input dataarray with a single step along the dimension that represent your frames (framedim).

mov_parallel = Movie(ds.air.chunk({'time':1}), vmin=230, vmax=310)

%%time
mov_parallel.save(
    'movie_parallel.mov',
    parallel=True,
    overwrite_existing=True,
)

Movie created at movie_parallel.mov
CPU times: user 38.8 s, sys: 1.46 s, total: 40.3 s
Wall time: 48.3 s

You can pass arguments to the dask .compute() call with parallel_compute_kwargs to tune for your particular setup.

%%time
mov_parallel.save(
    'movie_parallel_modified.mov',
    parallel=True,
    overwrite_existing=True,
    parallel_compute_kwargs=dict(scheduler="processes", num_workers=8)
)

Movie created at movie_parallel.mov
CPU times: user 4.84 s, sys: 249 ms, total: 5.09 s
Wall time: 33.6 s

Thats not bad, a 50% time saving (keeping in mind that the time needed for the ffmpeg call is included). We expect speedups to be even bigger when using higher resolution data.