Strategies

Classes

DefaultStrategy 

DefaultStrategy(data)

Bases: Strategy

Default strategy to be used when sampling/selecting nodes and checking whether nodes have responded or not

Strategy is: - select all node for each round - raise an error if one node does not answer - raise an error is one node returns an error

Parameters:

Name Type Description Default
data FederatedDataSet

Object that includes all active nodes and the meta-data of the dataset that is going to be used for federated training. Should be passed to super().__init__ to initialize parent class

 required
Source code in fedbiomed/researcher/strategies/default_strategy.py 
def __init__(self, data: FederatedDataSet):
    """ Constructor of Default Strategy

    Args:
        data: Object that includes all active nodes and the meta-data of the dataset that is going to be
            used for federated training. Should be passed to `super().__init__` to initialize parent class
    """

    super().__init__(data)

Functions

Strategy 

Strategy(data)

Default Strategy as Parent class. Custom strategy classes must inherit from this parent class.

    used for federated training.
Source code in fedbiomed/researcher/strategies/strategy.py 
def __init__(self, data: FederatedDataSet):
    """

    Args:
        data: Object that includes all active nodes and the meta-data of the dataset that is going to be
            used for federated training.
    """
    self._fds = data
    self._sampling_node_history = {}
    self._success_node_history = {}
    self._parameters = None

Functions