https://github.com/JelleAalbers/qsub_pywrap

Execute python functions as qsub jobs.

Simple job submission:

from qsub_pywrap import qsubwrap

def find_answer():
    return 42

jobname, output_pickle = qsubwrap(find_answer, queue='yourqueue')

When the job is done, the result will be in a pickle with filename output_pickle. Extra args/kwargs passed to qsubwrap will be passed to your function. (except if they're one of the options to qsubwrap, such as verbose, pickle_dir, messages_dir, ... see docstring).

Mapreduce functionality:

from qsub_pywrap import qsub_mapreduce

def find_partial_answer(i):
    return (i + 1.5)**2

def combine_results(x):
    return 1 + sum(x)

reducer_job, reducer_pickle = qsub_mapreduce(mapper=find_partial_answer,
                                             inputs=range(4),
                                             reducer=combine_results,
                                             queue='yourqueue')

Each call to mapper and the final call to reducer will happen in a separate job. The reducer job only starts when all of the mapper jobs are finished, and the results will be written to reducer_pickle. See the docstring for the various options of qsub_mapreduce, which e.g. allow you to pass extra arguments to the mapper and reducer.

A package with similar features is torque-submit https://pypi.python.org/pypi/torque-submit/0.0.3. This makes use of environment variables to pass data to jobs, and has quite a different interface. Documentation/docstrings are mostly absent.

If you just want to submit jobs that call a separate script or program, there are many, many alternatives: just google "qsub py" or "python qsub" and pick your favourite.

If you want detailed control over PBS jobs, you want pbs_python: https://oss.trac.surfsara.nl/pbs_python.