Skip to content

semipy.datasets.get_cifar

    semipy.datasets.cifar.get_cifar(name='cifar10',
                                    num_labelled=50000,
                                    num_unlabelled=0,
                                    valid_proportion=0.0,
                                    path='./data',
                                    augmentation=False,
                                    include_labelled=False) -> dict

Downloads one of the CIFAR dataset (CIFAR-10 or CIFAR-100) from torchvision and transforms it into a Semi-Supervised dataset (with labelled and unlabelled samples).

Parameters

  • name (str) - Name of the desired dataset. Should be one of ('cifar10', 'cifar100'). Default: 'cifar10'
  • num_labelled (int) - Number of labelled items to include in the dataset. Default: 50 000
  • num_unlabelled (int) - Number of unlabelled items to include in the dataset. Default: 0
  • valid_proportion (float) - Size of the validation set defined by a proportion with respect to the number of labelled items. Default: 0.0
  • path (str) - Path to download and store data samples. Default: './data'
  • augmentation (bool) - To use augmentation or not when sampling in the dataset. Default: False
  • include_labelled (bool) - To include labelled items into unlabelled set as additional data. Default: False