Data

dataset

The training and test datasets. The following options are available:

• HuggingFace: including all datasets from Hugging Face (requires dataset_name)
• Torchvision: including torchvision datasets such as MNIST, FashionMNIST, EMNIST, CIFAR10, CIFAR100, CelebA, or STL10 (requires dataset_name)
• CINIC10
• FEMNIST: Federated EMNIST
• TinyImageNet
• Purchase
• Texas

Torchvision configuration

When using the Torchvision datasource, specify dataset_name to choose the dataset class exposed by torchvision.datasets. Optional fields include:

• split_parameter: name of the constructor argument controlling the split (defaults to train or split when available).
• train_split, test_split, unlabeled_split: values passed to the split parameter for each subset. For boolean splits, strings such as "train" and "test" map to True and False.
• dataset_args / dataset_kwargs: positional or keyword arguments shared across all splits.
• train_args / train_kwargs (and the equivalents for test or unlabeled): per-split overrides.
• download: whether to trigger dataset downloads (defaults to true when supported by the selected dataset).
• For EMNIST, the balanced split is assumed by default; override dataset_kwargs = { split = "<variant>" } to select a different subset.
• For CelebA, attributes and identities are enabled by default; adjust dataset_kwargs.target_type when a different combination is required.

Sample Torchvision block

[data]
datasource = "Torchvision"
dataset_name = "MNIST"
download = true

dataset_kwargs = { root = "datasets" }
train_kwargs = { train = true }
test_kwargs = { train = false }

data_path

Where the dataset is located.

Default value: <base_path>/data

Note

For the CINIC10 dataset, the default is <base_path>/data/CINIC-10

For the TinyImageNet dataset, the default is <base_path>/data/tiny-imagenet-200

train_path

Where the training dataset is located.

test_path

Where the test dataset is located.

sampler

How to divide the entire dataset to the clients. The following options are available:

• iid

• noniid Could have concentration attribute to specify the concentration parameter in the Dirichlet distribution

  concentration

  If the sampler is noniid, the concentration parameter for the Dirichlet distribution can be specified.

  Default value: 1

• orthogonal Each institution's clients have data of different classes. Could have institution_class_ids and label_distribution attributes

  institution_class_ids

  If the sampler is orthogonal, the indices of classes of local data of each institution's clients can be specified. e.g., 0, 1; 2, 3 (the first institution's clients only have data of class #0 and #1; the second institution's clients only have data of class #2 and #3).

  label_distribution

  If the sampler is orthogonal, the class distribution of every client's local data can be specified. The value should be iid or noniid.

  Default value: iid

• mixed Some data are iid, while others are non-iid. Must have non_iid_clients attributes

  non_iid_clients

  If the sampler is mixed, the indices of clients whose datasets are non-i.i.d. need to be specified. Other clients' datasets are i.i.d.

testset_sampler

How the test dataset is sampled when clients test locally. Any sampler type is valid.

Note

Without this parameter, the test dataset on either the client or the server is the entire test dataset of the datasource.

random_seed

The random seed used to sample each client's dataset so that experiments are reproducible.

partition_size

The number of samples in each client's dataset.

testset_size

The number of samples in the server's test dataset when server-side evaluation is conducted.