Skip to main content

base

Base classes for all algorithms.

Each module in this package defines a single algorithm.

Attributes: registry: A read-only dictionary of algorithm factory names to their implementation classes.

Classes​

AlgorithmDecoratorMetaClass​

class AlgorithmDecoratorMetaClass(*args, **kwargs):

Decorates the __init__, initialise and run algorithm methods.

Ancestors​

  • bitfount.hooks.BaseDecoratorMetaClass
  • builtins.type

Subclasses​

  • types.AbstractAlgorithmDecoratorMetaClass

Static methods​

decorator​

def decorator(f: Callable) ‑> collections.abc.Callable:

Hook and federated error decorators.

do_decorate​

def do_decorate(attr: str, value: Any) ‑> bool:

Checks if an object should be decorated.

Only the init, initialise and run methods should be decorated.

BaseAlgorithmFactory​

class BaseAlgorithmFactory(**kwargs: Any):

Base algorithm factory from which all other algorithms must inherit.

Attributes

  • class_name: The name of the algorithm class.

Ancestors​

  • abc.ABC
  • bitfount.federated.roles._RolesMixIn
  • bitfount.types._BaseSerializableObjectMixIn
  • typing.Generic

Subclasses​

Variables​

BaseInitialSetupAlgorithm​

class BaseInitialSetupAlgorithm(**kwargs: Any):

Base class for algorithms that perform initial setup.

Provides a common base implementation for algorithms that need to run setup operations before batching begins.

Ancestors​

Methods​

initialise​

def initialise(    self,    *,    datasource: BaseSource,    task_id: str,    data_splitter: Optional[DatasetSplitter] = None,    pod_dp: Optional[DPPodConfig] = None,    pod_identifier: Optional[str] = None,    **kwargs: Any,) ‑> None:

Inherited from:

BaseWorkerAlgorithm.initialise :

Initialises the algorithm.

This method is only called once regardless of the number of batches in the task.

note

This method must call the initialise_data method.

initialise_data​

def initialise_data(    self, datasource: BaseSource, data_splitter: Optional[DatasetSplitter] = None,) ‑> None:

Inherited from:

BaseWorkerAlgorithm.initialise_data :

Initialises the algorithm with data.

This method will be called once per task batch. It is expected that algorithms will override this method to initialise their data in the required way.

note

This is called by the initialise method and should not be called directly by the algorithm or protocol.

setup_run​

def setup_run(self, **kwargs: Any) ‑> None:

Run setup operations before batching begins.

This method will be called by protocols tagged with InitialSetupProtocol before any batching occurs.

BaseModellerAlgorithm​

class BaseModellerAlgorithm(**kwargs: Any):

Modeller side of the algorithm.

Ancestors​

  • bitfount.federated.algorithms.base._BaseAlgorithm
  • abc.ABC

Subclasses​

  • NoResultsModellerAlgorithm
  • ResultsOnlyModellerAlgorithm
  • bitfount.federated.algorithms.hugging_face_algorithms.base._HFModellerSide
  • bitfount.federated.algorithms.model_algorithms.base._BaseModellerModelAlgorithm
  • bitfount.federated.algorithms.s3_upload_algorithm._ModellerSide

Methods​

initialise​

def initialise(self, *, task_id: str, **kwargs: Any) ‑> None:

Initialise the algorithm.

BaseNonModelAlgorithmFactory​

class BaseNonModelAlgorithmFactory(*, datastructure: DataStructure, **kwargs: Any):

Base factory for algorithms not involving an underlying model.

Arguments

  • datastructure: The data structure to use for the algorithm.
  • ****kwargs**: Additional keyword arguments.

Attributes

  • datastructure: The data structure to use for the algorithm.

Ancestors​

Subclasses​

Variables​

Methods​

modeller​

def modeller(self, *, context: ProtocolContext, **kwargs: Any) ‑> ~T_ModellerSide:

Modeller-side of the algorithm.

worker​

def worker(self, *, context: ProtocolContext, **kwargs: Any) ‑> ~T_WorkerSide:

Worker-side of the algorithm.

BaseWorkerAlgorithm​

class BaseWorkerAlgorithm(**kwargs: Any):

Worker side of the algorithm.

Ancestors​

  • bitfount.federated.algorithms.base._BaseAlgorithm
  • abc.ABC

Subclasses​

  • BaseInitialSetupAlgorithm
  • bitfount.federated.algorithms.csv_report_algorithm._WorkerSide
  • BaseEHRWorkerAlgorithm
  • bitfount.federated.algorithms.ehr.image_selection_algorithm._WorkerSide
  • bitfount.federated.algorithms.filtering_algorithm._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_image_classification._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_image_segmentation._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_perplexity._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_text_classification._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_text_generation._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.timm_fine_tuning._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.timm_inference._WorkerSide
  • bitfount.federated.algorithms.model_algorithms.base._BaseWorkerModelAlgorithm
  • bitfount.federated.algorithms.ophthalmology.bscan_image_and_mask_generation_algorithm._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.csv_report_generation_ophth_algorithm._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.etdrs_calculation_algorithm._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.fluid_volume_calculation_algorithm._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.fovea_coordinates_algorithm._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.ga_trial_calculation_algorithm_base._BaseWorkerSideWithFovea
  • bitfount.federated.algorithms.ophthalmology.ga_trial_calculation_algorithm_base._WorkerSide
  • BaseGATrialInclusionWorkerAlgorithm
  • bitfount.federated.algorithms.ophthalmology.ga_trial_pdf_algorithm_base._BasePDFWorkerSide
  • bitfount.federated.algorithms.ophthalmology.reduce_csv_algorithm_charcoal._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.simple_csv_algorithm._WorkerSide
  • bitfount.federated.algorithms.private_sql_query._WorkerSide
  • bitfount.federated.algorithms.s3_upload_algorithm._WorkerSide
  • bitfount.federated.algorithms.sql_query._WorkerSide

Methods​

initialise​

def initialise(    self,    *,    datasource: BaseSource,    task_id: str,    data_splitter: Optional[DatasetSplitter] = None,    pod_dp: Optional[DPPodConfig] = None,    pod_identifier: Optional[str] = None,    **kwargs: Any,) ‑> None:

Initialises the algorithm.

This method is only called once regardless of the number of batches in the task.

note

This method must call the initialise_data method.

initialise_data​

def initialise_data(    self, datasource: BaseSource, data_splitter: Optional[DatasetSplitter] = None,) ‑> None:

Initialises the algorithm with data.

This method will be called once per task batch. It is expected that algorithms will override this method to initialise their data in the required way.

note

This is called by the initialise method and should not be called directly by the algorithm or protocol.

FinalStepAlgorithm​

class FinalStepAlgorithm():

Tagging interface for algorithms that contain a final "reduce" step.

These algorithms have steps that can be operated batch-wise (normal run()) followed by step(s) at the end that cannot be executed batch-wise but instead require access to outputs from all batch steps (reduce step(s)).

Subclasses​

  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_image_classification._WorkerSide
  • bitfount.federated.algorithms.hugging_face_algorithms.hugging_face_image_segmentation._WorkerSide
  • bitfount.federated.algorithms.ophthalmology.csv_report_generation_ophth_algorithm._WorkerSide

Methods​

run_final_step​

def run_final_step(self, *, context: ProtocolContext, **kwargs: Any) ‑> Any:

Execute the final step without datasource access.

This method is called after all batch processing is complete and should not access the datasource directly.

Arguments

  • context: The protocol context containing runtime information for the protocol.
  • ****kwargs**: Additional context like accumulated results, task_id, etc.

Returns Final processed results

InitialSetupAlgorithm​

class InitialSetupAlgorithm(*args, **kwargs):

Protocol defining the interface for algorithms that perform initial setup.

Any algorithm used with InitialSetupProtocol must implement this interface.

Ancestors​

Methods​

setup_run​

def setup_run(self, **kwargs: Any) ‑> None:

Run setup operations before batching begins.

This method will be called by protocols tagged with InitialSetupProtocol before any batching occurs.

NoResultsModellerAlgorithm​

class NoResultsModellerAlgorithm(    log_message: Optional[str] = None,    save_path: Optional[Union[os.PathLike, str]] = None,    **kwargs: Any,):

Modeller side of any algorithm that does not return any results.

Arguments

  • log_message: A message to log when the algorithm is run on the modeller side.
  • save_path: The path the modeller results should be saved to, if any.
  • ****kwargs**: Additional keyword arguments.

Ancestors​

Methods​

initialise​

def initialise(self, *, task_id: str, **kwargs: Any) ‑> None:

Nothing to initialise here.

run​

def run(self, results: Mapping[str, Any]) ‑> None:

Modeller side just logs the log message.

ResultsOnlyModellerAlgorithm​

class ResultsOnlyModellerAlgorithm(**kwargs: Any):

Modeller side of any algorithm that only returns results.

Ancestors​

Methods​

initialise​

def initialise(self, *, task_id: str, **kwargs: Any) ‑> None:

Nothing to initialise here.

run​

def run(self, results: Mapping[str, Any]) ‑> dict[str, typing.Any]:

Modeller side just returns the results as a dictionary.