Swarm - Define your decentralized algorithm¶

class manta.swarm.Swarm¶

This class defines a decentralized algorithm which must be sent to a cluster

globals¶

Initial global values

Type:: Dict[str, Any]

execute() → ModuleID¶

This method must be defined by the User

Returns:: Module ID which is the last module executed in the graph before iterating
Return type:: ModuleID

Examples

>>> def execute(self):
...     """
...     Generation of the task graph
...
...     +--------+     +------------+     +-----------+ if has_converged
...     | Worker | --> | Aggregator | --> | Scheduler | ----------------> END PROGRAM
...     +--------+     +------------+     +-----------+
...         |                                   | else
...         +--<<<----------<<<----------<<<----+
...     """
...     m = self.worker()
...     m = self.aggregator(m)
...     return self.scheduler(m)

set_global(tag: str, value: dict)¶

Set a global value in the swarm

Examples

>>> self.set_global(
...     "hyperparameters",
...     {
...         "epochs": 1,
...         "batch_size": 32,
...         "loss": "CrossEntropyLoss",
...         "loss_params": {},
...         "optimizer": "SGD",
...         "optimizer_params": {"lr": 0.01, "momentum": 0.9},
...     },
... )