MODULE REFERENCE

torch_autoencoder

anomx.models.reconstruction.torch_autoencoder

PyTorch Lightning autoencoder model (requires the torch extra).

TorchAutoencoderModel

TorchAutoencoderModel(config: dict[str, Any] | None = None)

Bases: NormalityModel, Reconstructive

Small fully-connected autoencoder trained with PyTorch Lightning.

Reconstruction error per observation lands in model_score. The component exists to establish the torch/lightning integration pattern; the network is intentionally tiny.

Signature parameters
ParameterTypeDefault
configdict[str, Any] | NoneNone

Members

component_keyattribute
component_key = 'torch_autoencoder'
component_iconattribute
signatureattribute
signature = ModelSignature(structures={DataStructure.TABULAR, DataStructure.TEMPORAL_SEQUENCE}, modalities={Modality.NUMERIC})
component_nameattribute
component_name = 'Torch Autoencoder'
component_default_configattribute
component_default_config = {'feature_columns': [], 'hidden_size': 8, 'latent_size': 2, 'max_epochs': 5}
component_config_schemaattribute
component_config_schema = {'feature_columns': {'type': 'array'}, 'hidden_size': {'type': 'integer'}, 'latent_size': {'type': 'integer'}, 'max_epochs': {'type': 'integer'}}
capabilitiesattribute
capabilities: ClassVar[ComponentCapabilities] = ComponentCapabilities(supports_sequences=True)
fitmethod
fit(data: Any) -> None
Signature parameters
ParameterTypeDefault
dataAnyrequired

Returns None

predictmethod
predict(data: Any) -> pd.DataFrame
Signature parameters
ParameterTypeDefault
dataAnyrequired

Returns pd.DataFrame

savemethod
save(path: str) -> None
Signature parameters
ParameterTypeDefault
pathstrrequired

Returns None

loadmethod
load(path: str) -> TorchAutoencoderModel
Signature parameters
ParameterTypeDefault
pathstrrequired

Returns TorchAutoencoderModel