Function - get_dataloader

Contents

Function - get_dataloader#

openfl.utilities.dataloading.get_dataloader(plan, prefer_minimal=False, input_shape=None, collaborator_index=0)[source]#

Get dataloader instance from plan

NOTE: if prefer_minimal is False, cwd must be the workspace directory because we need to construct dataloader from actual collaborator data path with actual data present.

Parameters:

  • plan (Plan) – plan object linked with the dataloader

  • prefer_minimal (bool ?) – prefer to use MockDataLoader which can be used to more easily instantiate task_runner without any initial data. Default to False.

  • input_shape (list | dict ?) – input_shape denoted by list notation [a,b,c, …] or in case of multihead models, dict object with individual layer keys such as {“input_0”: [a,b,…], “output_1”: [x,y,z, …]} Defaults to None.

  • collaborator_index (int ?) – which collaborator should be used for initializing dataloader among collaborators specified in plan/data.yaml. Defaults to 0.

Returns:

DataLoader instance

Return type:

data_loader (DataLoader)

Contents