synthcity.plugins.core.dataset module

class ConditionalDataset(data: torch.utils.data.dataset.Dataset, cond: Optional[torch.Tensor] = None)

Bases: torch.utils.data.dataset.Dataset

Helper dataset for wrapping existing datasets with custom tensors

Parameters

• data – torch.Dataset

• cond – Optional Tensor

class FlexibleDataset(data: torch.utils.data.dataset.Dataset, transform: Optional[torch.nn.modules.module.Module] = None, indices: Optional[list] = None)

Bases: torch.utils.data.dataset.Dataset

Helper dataset wrapper for post-processing or transforming another dataset. Used for controlling the image sizes for the synthcity models.

The class supports adding custom transforms to existing datasets, and to subsample a set of indices.

Parameters

• data – torch.Dataset

• transform – An optional list of transforms

• indices – An optional list of indices to subsample

filter_indices(indices: List[int]) → synthcity.plugins.core.dataset.FlexibleDataset

labels() → numpy.ndarray

numpy() → Tuple[numpy.ndarray, numpy.ndarray]

shape() → Tuple

tensors() → Tuple[torch.Tensor, torch.Tensor]

class NumpyDataset(X: numpy.ndarray, y: numpy.ndarray)

Bases: torch.utils.data.dataset.Dataset

Helper class for wrapping Numpy arrays in torch Datasets :param X: np.ndarray :param y: np.ndarray

class TensorDataset(images: torch.Tensor, targets: Optional[torch.Tensor])

Bases: torch.utils.data.dataset.Dataset

Helper dataset for wrapping existing tensors

Parameters

• images – Tensor

• targets – Tensor

labels() → Optional[numpy.ndarray]