Compute Plan Builder

class substrafl.compute_plan_builder.ComputePlanBuilder(*args, **kwargs)

Bases: ABC

Base compute plan builder to be inherited from SubstraFL compute plans.

All child class arguments need to be passed to it through its args and kwargs in order to use them when instantiating it as a RemoteStruct in each process.

Example

class MyComputePlan(ComputePlanBuilder):
    def __init__(self, custom_arg, my_custom_kwargs="value"):
        super().__init__(custom_arg, my_custom_kwargs=my_custom_kwargs)
abstract build_compute_plan(train_data_nodes: List[TrainDataNodeProtocol] | None, aggregation_node: List[AggregationNodeProtocol] | None, evaluation_strategy: EvaluationStrategy | None, num_rounds: int | None, clean_models: bool | None = True) → None

Build the compute plan to be executed. All arguments are optional and will be feed within the execute_experiment() function.

Parameters:

  • train_data_nodes (List[TrainDataNodeProtocol]) – list of the train organizations

  • aggregation_node (Optional[AggregationNodeProtocol]) – aggregation node, necessary for centralized strategy, unused otherwise

  • evaluation_strategy (Optional[EvaluationStrategy]) – evaluation strategy to follow for testing models.

  • num_rounds (int) – Number of times to repeat the compute plan sub-graph (define in perform round). It is useful in recurring graphs, but can be ignored in other cases.

  • clean_models (bool) – Clean the intermediary models on the Substra platform. Set it to False if you want to download or re-use intermediary models. This causes the disk space to fill quickly so should be set to True unless needed. Defaults to True.

Returns:

None

Return type:

None

abstract load_local_state(path: Path) → Any

Executed at the beginning of each step of the computation graph to load on each organization the previously saved local state.

Parameters:

path (pathlib.Path) – The path where the previous local state has been saved.

Returns:

The loaded element.

Return type:

Any

abstract save_local_state(path: Path) → None

Executed at the end of each step of the computation graph to save the local state locally on each organization.

Parameters:

path (pathlib.Path) – The path where the previous local state has been saved.

Returns:

None

Return type:

None