Fetching contributors…
Cannot retrieve contributors at this time
47 lines (38 sloc) 1.6 KB
Test the NetworkPPScheduler.
This script only works on *nix systems. On Windows systems you can create
the slave servers and the pp scheduler instance manually and then use it with
MDP's standard PPScheduler.
import numpy as np
import mdp
import mdp.parallel as parallel
# start a single pp slave server on localhost with support for two cores
remote_slaves = [("localhost", 2)]
# IMPORTANT: modify these paths to make sure that MDP and the
# module (included in pp) can be importet on the slave machines.
sys_paths = ["/home/niko/win_workspace/mdp-toolkit",
"/home/niko/win_workspace/Parallel\ Python/src"]
def main():
# create the scheduler, which in turn starts the remote slave servers
with parallel.pp_support.NetworkPPScheduler(
verbose=False) as scheduler:
# test simple tasks
for i in range(30):
scheduler.add_task(i, parallel.SqrTestCallable())
results = scheduler.get_results()
# check result
results = np.array(results)
assert np.all(results[:6] == np.array([0,1,4,9,16,25]))
print "simple test done, now testing with MDP flow..."
# simple test with MDP flow
flow = mdp.parallel.ParallelFlow([mdp.nodes.PCANode()])
data = [np.random.random([50, 10]) for _ in range(5)]
flow.train([data], scheduler)
print "done."
if __name__ == "__main__":