Caffe2 - Python API
A deep learning, cross platform ML framework
Public Member Functions | Static Public Attributes | List of all members
caffe2.python.task.TaskGroup Class Reference
Inheritance diagram for caffe2.python.task.TaskGroup:

Public Member Functions

def __init__ (self, workspace_type=None)
 
def add (self, task)
 
def tasks (self)
 
def num_registered_tasks (self)
 
def used_nodes (self)
 
def report_step (self, step=None, node=None, interval_ms=1000)
 
def report_net (self, net=None, node=None, report_interval=5)
 
def tasks_by_node (self, node_remap=None)
 
def to_task (self, node=None)
 
def workspace_type (self)
 

Static Public Attributes

string LOCAL_SETUP = 'local_setup'
 

Detailed Description

Context that gathers tasks which will run concurrently, potentially on
multiple nodes. All tasks in the same node will share the same workspace
and thus can share blobs, while tasks running in different nodes won't
be able to directly share data.

All tasks of the task group will start concurrently, and the task group
will finish execution when the last task of the group finishes.

Example:
    # supose that s1 ... s5 are execution steps or nets.
    with TaskGroup() as tg:
        # these tasks go to default node 'local'
        Task(step=s1)
        Task(step=s2)

        with Node('n2'):
            Task(step=s3)
        with Node('n1'):
            Task(step=s4)
        with Node('n2'):
            Task(step=s5)

    # this will run all steps in parallel.
    # s1 and s2 will run at default node 'local'
    # s3 and s5 will run at node 'n2'
    # s4 will run at node 'n1'
    session.run(tg)

Definition at line 174 of file task.py.

Member Function Documentation

def caffe2.python.task.TaskGroup.report_net (   self,
  net = None,
  node = None,
  report_interval = 5 
)
DEPRECATED. Use report_step instead.

Definition at line 261 of file task.py.

def caffe2.python.task.TaskGroup.report_step (   self,
  step = None,
  node = None,
  interval_ms = 1000 
)
Add a "report step" to this TaskGroup. This step will run repeatedly
every `interval_ms` milliseconds for the duration of the TaskGroup
execution on each of the nodes. It is guaranteed that this step
will be run at least once after every Task in the node has finished.

Definition at line 250 of file task.py.


The documentation for this class was generated from the following file: