nexuml_library.scenarios.asd.dcase_conv_ae_baseline¶
DCASE convolutional autoencoder scenarios.
dcase_conv_ae(
machine_types: list[str] | None = None,
machine_specs: list[MachineSpec] | None = None,
download: bool = False,
n_mels: int = 128,
hop_length: int = 512,
clip_num_samples: int = 160000,
latent_dim: int = 64,
channel_schedule: list[int] | None = None,
lr: float = 0.001,
batch_size: int = 16,
max_epochs: int = 25,
) -> ScenarioSpec
DCASE convolutional autoencoder anomaly workflow.
Machine year and data_type come from the built-in DCASE catalog when
machine_types is given. Use machine_specs for explicit multi-year
control. time_frames is computed from clip_num_samples / hop_length.
Returns:
| Name | Type | Description |
|---|---|---|
ScenarioSpec |
ScenarioSpec
|
Assembled scenario with pipeline, training, data and evaluation. |
dcase_conv_cvae(
machine_types: list[str] | None = None,
machine_specs: list[MachineSpec] | None = None,
download: bool = False,
n_mels: int = 128,
hop_length: int = 512,
clip_num_samples: int = 160000,
encoder_dim: int = 128,
latent_dim: int = 32,
beta: float = 0.001,
channel_schedule: list[int] | None = None,
lr: float = 0.001,
batch_size: int = 16,
max_epochs: int = 25,
) -> ScenarioSpec
DCASE convolutional variational autoencoder anomaly workflow.
Machine year and data_type come from the built-in DCASE catalog when
machine_types is given. time_frames is computed automatically.
Returns:
| Name | Type | Description |
|---|---|---|
ScenarioSpec |
ScenarioSpec
|
Assembled scenario with pipeline, training, data and evaluation. |