def __init__(self, num_workers: int, host_ip):
"""Start Rabit tracker. The workers connect to this tracker to share
their results."""
self._num_workers = num_workers
self.env = {"DMLC_NUM_WORKER": self._num_workers}
self.rabit_tracker = xgb.RabitTracker(hostIP=host_ip,
nslave=self._num_workers)
def _start_rabit_tracker(num_workers: int):
"""Start Rabit tracker. The workers connect to this tracker to share
their results."""
host = get_node_ip_address()
env = {"DMLC_NUM_WORKER": num_workers}
rabit_tracker = xgb.RabitTracker(hostIP=host, nslave=num_workers)
# Get tracker Host + IP
env.update(rabit_tracker.slave_envs())
rabit_tracker.start(num_workers)
# Wait until context completion
thread = Thread(target=rabit_tracker.join)
thread.daemon = True
thread.start()
return env
def _start_rabit_tracker(num_workers: int):
"""Start Rabit tracker. The workers connect to this tracker to share
their results."""
# TODO (hme): Cleanup thread and tracker after training.
host = systems_utils.get_private_ip()
env = {"DMLC_NUM_WORKER": num_workers}
rabit_tracker = xgb.RabitTracker(hostIP=host, nslave=num_workers)
# Get tracker Host + IP
env.update(rabit_tracker.slave_envs())
rabit_tracker.start(num_workers)
# Wait until context completion
thread = Thread(target=rabit_tracker.join)
thread.daemon = True
thread.start()
return env
def __init__(self, num_workers: int, host_ip):
self._num_workers = num_workers
self.env = {"DMLC_NUM_WORKER": self._num_workers}
self.rabit_tracker = xgb.RabitTracker(hostIP=host_ip,
nslave=self._num_workers)