Dataset generator for Semi-supervised Learning

In this section, we consider the dataset generator function for performing experiments in both standard and robust semi-supervised learning scenarios.

Dataset Builder

cords.utils.data.datasets.SSL.builder.gen_dataloader(root, dataset, validation_split, cfg, logger=None)[source]

generate train, val, and test dataloaders

Parameters

  • root (str) – root directory

  • dataset (str) – dataset name, [‘cifar10’, ‘cifar100’, ‘svhn’, ‘stl10’]

  • validation_split (bool) – if True, return validation loader. validation data is made from training data

  • cfg (argparse.Namespace or something) –

  • logger (logging.Logger) –

cords.utils.data.datasets.SSL.builder.gen_dataset(root, dataset, validation_split, cfg, logger=None)[source]

generate train, val, and test datasets

Parameters

  • root (str) – root directory in which data is present or needs to be downloaded

  • dataset (str) – dataset name, Existing dataset choices: [‘cifar10’, ‘cifar100’, ‘svhn’, ‘stl10’, ‘cifarOOD’, ‘mnistOOD’, ‘cifarImbalance’]

  • validation_split (bool) – if True, return validation loader. We use 10% random split of training data as validation data

  • cfg (argparse.Namespace or dict) – Dictionary containing necessary arguments for generating the dataset

  • logger (logging.Logger) – Logger class for logging the information