`nexuml_library.scenarios.data.dcase`¶

nexuml_library.scenarios.data.dcase ¶

DCASE Task 2 data specification builders.

MachineSpec ¶

Bases: BaseModel

Identifies a DCASE machine type together with the year it was released.

The year determines which dataset subdirectory the files live in (e.g. year=2023 → dcase2023t2/). data_type is either "dev" (labelled development set) or "eval" (evaluation set).

key `property` ¶

key: str

Unique machine identity across years/data splits.

dcase_machine_specs ¶

dcase_machine_specs(
    machine_types: list[str] | None = None,
    years: list[int] | None = None,
    data_types: list[str] | None = None,
    dedupe_machine_types: bool = False,
) -> list[MachineSpec]

Return known DCASE machine specs from the static scenario catalog.

When dedupe_machine_types is true, keep at most one spec per raw machine type, preferring eval over dev and newer years over older years.

dcase_data ¶

dcase_data(
    data_root: str | Path = "DCASET2",
    machine_types: list[str] | None = None,
    machine_specs: list[MachineSpec] | None = None,
    train_machine_specs: list[MachineSpec] | None = None,
    test_machine_specs: list[MachineSpec] | None = None,
    years: list[int] | None = None,
    data_types: list[str] | None = None,
    dedupe_machine_types: bool = False,
    download: bool = False,
    sample_rate: int = 16000,
    clip_num_samples: int = 160000,
    batch_size: int = 64,
    num_workers: int = 4,
    validate_machine_specs: bool = True,
) -> DataSpec

DCASE Task 2 anomaly detection data spec.

Pass machine_specs for explicit multi-year control, or machine_types to filter the built-in DCASE machine catalog.

Set NEXUML_DATA_ROOT so scenario files can use logical dataset paths without cluster-specific paths in code.

Parameters:

Name	Type	Description	Default
`data_root`	`str \| Path`	Logical root. Resolved via `NEXUML_DATA_ROOT` if the path does not exist as-is.	`'DCASET2'`
`machine_types`	`list[str] \| None`	Names of machine types to keep from the catalog.	`None`
`machine_specs`	`list[MachineSpec] \| None`	Explicit per-machine year + data_type (multi-year) for both fit and test.	`None`
`train_machine_specs`	`list[MachineSpec] \| None`	Explicit machine specs used only for fit.	`None`
`test_machine_specs`	`list[MachineSpec] \| None`	Explicit machine specs used only for test.	`None`
`years`	`list[int] \| None`	Optional year filter.	`None`
`data_types`	`list[str] \| None`	Optional data type filter, e.g. ["eval"] for additional/eval machines.	`None`
`dedupe_machine_types`	`bool`	Keep at most one spec per raw machine type, preferring eval/newer year.	`False`
`download`	`bool`	Download/extract missing Zenodo zips from the DCASE manifest.	`False`
`sample_rate`	`int`	Audio sample rate in Hz.	`16000`
`clip_num_samples`	`int`	Samples per audio clip.	`160000`
`batch_size`	`int`	DataLoader batch size.	`64`
`num_workers`	`int`	DataLoader workers.	`4`
`validate_machine_specs`	`bool`	Validate explicit machine_specs against the catalog. Set to False to allow custom/experimental machines not in the catalog.	`True`

Returns:

Name	Type	Description
`DataSpec`	`DataSpec`	DCASE dataset specification with fit and test splits.

nexuml_library.scenarios.data.dcase¶