dataeval.types.ExtractorInfo¶

class dataeval.types.ExtractorInfo¶

Descriptor for a feature extractor as used to produce embeddings.

Denormalized form of DataEval-Flow’s per-model extractor config: the underlying model and the preprocessor pipeline are inlined rather than referenced by name, so the sidecar is self-contained.

type¶

Extractor class name (e.g. "OnnxExtractor", "TorchExtractor", "FlattenExtractor", "BoVWExtractor", "ClassifierUncertaintyExtractor").

Type:¶

str

model¶

Underlying model artifact. None for extractors with no model (e.g. FlattenExtractor).

Type:¶

ModelInfo or None

transforms¶

Ordered preprocessing pipeline applied before model inference.

Type:¶

list of TransformInfo

params¶

Extractor-specific runtime parameters (e.g. output_name, layer_name, flatten, device, batch_size).

Type:¶

dict