ClearML/clearml
1
0
Fork 0
mirror of https://github.com/clearml/clearml synced 2025-03-03 18:52:12 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Improve example documentation

Browse Source
This commit is contained in:
allegroai 2020-05-24 15:47:52 +03:00
parent 9e0a5a36fb
commit 342fac71e2
2 changed files with 20 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
examples/distributed/example_subprocess.py
View File

@ -9,6 +9,7 @@ from argparse import ArgumentParser
from trains import Task
# fake data for us to "process"
data = (
['a', '2'], ['b', '4'], ['c', '6'], ['d', '8'],
['e', '1'], ['f', '3'], ['g', '5'], ['h', '7'],
@ -39,17 +40,18 @@ def mp_handler(use_subprocess):
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('--num_workers', help='integer value', type=int, default=3)
parser.add_argument('--counter', help='integer value', type=int, default=-1)
parser.add_argument('--use_subprocess', help='integer value', type=int, default=1)
# this argument we will not be logging, see below Task.init
parser.add_argument('--counter', help='integer value', type=int, default=-1)
args = parser.parse_args()
print(os.getpid(), 'ARGS:', args)
# We have to initialize the task in the master process,
# it will make sure that any sub-process calling Task.init will get the master task object
# notice that we exclude the counter argument, so we can launch multiple sub-processes with trains-agent
# otherwise, the counter will always be set to the original value
task = Task.init('examples', 'POpen example', auto_connect_arg_parser={'counter': False})
# notice that we exclude the `counter` argument, so we can launch multiple sub-processes with trains-agent
# otherwise, the `counter` will always be set to the original value.
task = Task.init('examples', 'Popen example', auto_connect_arg_parser={'counter': False})
# we can connect multiple dictionaries, each from different process, as long as the keys have different names
param = {'args_{}'.format(args.num_workers): 'some value {}'.format(args.num_workers)}
@ -59,6 +61,8 @@ if __name__ == '__main__':
counter = args.num_workers if args.counter < 0 else args.counter
p = None
# launch sub-process, every subprocess will launch the next in the chain, until we launch them all.
# We could also launch all of them here, but that would have been to simple for us J
if counter > 0:
cmd = [sys.executable, sys.argv[0],
'--counter', str(counter - 1),
@ -67,8 +71,13 @@ if __name__ == '__main__':
print(cmd)
p = subprocess.Popen(cmd, cwd=os.getcwd())
# the actual "processing" is done here
mp_handler(args.use_subprocess)
print('Done logging')
# wait for the process we launched
# this means every subprocess will be waiting for the process it launched and
# the master process will exit after all of them are completed
if p and counter > 0:
p.wait()
print('Exiting')

8
examples/distributed/example_torch_distributed.py
View File

@ -143,10 +143,16 @@ if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument('--nodes', help='number of nodes', type=int, default=10)
parser.add_argument('--workers_in_node', help='number of workers per node', type=int, default=3)
# this argument we will not be logging, see below Task.init
parser.add_argument('--rank', help='current rank', type=int)
args = parser.parse_args()
task = Task.init("examples", "test torch distributed")
# We have to initialize the task in the master process,
# it will make sure that any sub-process calling Task.init will get the master task object
# notice that we exclude the `rank` argument, so we can launch multiple sub-processes with trains-agent
# otherwise, the `rank` will always be set to the original value.
task = Task.init("examples", "test torch distributed", auto_connect_arg_parser={'rank': False})
if os.environ.get('MASTER_ADDR'):
dist.init_process_group(backend='gloo', rank=args.rank, world_size=args.nodes)