simet.schemas.loader

simet.schemas.loader

LoaderSchema dataclass

LoaderSchema(real_provider, synth_provider, provider_transform, feature_extractor)

Schema describing how to build a DatasetLoader (providers + transforms + FE).

Encapsulates the components needed to construct the data loading and feature-extraction pipeline: one provider for real data, one for synthetic, a transform used for provider/datalaoder preprocessing, and the feature extractor configuration.

Attributes:

Name	Type	Description
real_provider	ProviderSchema	Configuration for the real dataset provider.
synth_provider	ProviderSchema	Configuration for the synthetic dataset provider.
provider_transform	TransformSchema	Transform applied when constructing the provider-backed datasets and their dataloaders (i.e., preprocessing seen by the feature extractor).
feature_extractor	FeatureExtractorSchema	Feature extractor selection/config (e.g., "inception_v3").

Example

cfg = LoaderSchema( ... real_provider=ProviderSchema(type="local_binary", data_path="data/real"), ... synth_provider=ProviderSchema(type="local_binary", data_path="data/synth"), ... provider_transform=TransformSchema(type="inception"), ... feature_extractor=FeatureExtractorSchema(type="inception_v3"), ... ) cfg.real_provider.type 'local_binary'

Notes

• real_provider and synth_provider should be symmetric (same structure), differing mainly in their data paths.
• The provider_transform should match the expectations of the feature_extractor to avoid feature drift (e.g., input size/normalization).