FedDataset

FedDataset#

class fl_sim.data_processing.FedDataset[源代码]#

基类：ReprMixin, CitationMixin, ABC

Base class for all federated datasets.

Methods that have to be implemented by subclasses:

  • get_dataloader

  • _preload

  • load_partition_data

  • load_partition_data_distributed

  • evaluate

Properties that have to be implemented by subclasses:

  • url

  • candidate_models

  • doi

abstract property candidate_models: Dict[str, Module]#

A set of candidate models.

property data_parts: List[str]#

Data part names.

abstract property doi: str | List[str]#

DOI(s) related to the dataset.

download_if_needed() → None[源代码]#

Download data if needed.

abstract evaluate(probs: Tensor, truths: Tensor) → Dict[str, float][源代码]#

Evaluation using predictions and ground truth.

extra_repr_keys() → List[str][源代码]#

Extra keys for __repr__() and __str__().

abstract get_dataloader(train_bs: int, test_bs: int, client_idx: int | None = None) → Tuple[DataLoader, DataLoader][源代码]#

Get dataloader for client client_idx or get global dataloader.

abstract load_partition_data(batch_size: int | None = None) → tuple[源代码]#

Partition data into all local clients.

abstract load_partition_data_distributed(process_id: int, batch_size: int | None = None) → tuple[源代码]#

Get local dataloader at client process_id or get global dataloader

property mirrors: List[Tuple[str, str | None]]#

Download mirrors as (url, md5) pairs, tried in order.

The first mirror defaults to url without checksum; subclasses can override this to add fallback mirrors and MD5 checksums.

abstract property url: str#

URL for downloading the dataset.

目录