ageas.Unit

class ageas.Unit(tail: str, unit_type: str, config: dict, accelerator: str = 'cpu')

Bases: object

Single classifier model bundled with its configuration and runtime state.

The unit stores everything needed to reproduce, train, and inspect one model: the classifier type, the configuration dictionary, the chosen accelerator, the trained model weights, the trainer object, and the accumulated reports from each Ageas operation.

Variables:

  • tail – Short name used to disambiguate units sharing the same type.

  • type – Model type key (one of SUPPORTED_TYPES).

  • config – Hyper-parameter dictionary for the classifier and trainer.

  • accelerator – Current accelerator string ('cpu' or 'cuda').

  • device – GPU device count / index resolved at spike time.

  • model – Trained classifier instance (None until after a sortie).

  • trainer – Trainer object (None until after a sortie).

  • report – Per-operation metric records accumulated during selection.

__init__(tail: str, unit_type: str, config: dict, accelerator: str = 'cpu') → None

Initialize a Unit.

Parameters:

  • tail – Short tail name used to disambiguate units that share the same model type (e.g. 'unit1', 'unit2').

  • unit_type – Model type key. Must be one of SUPPORTED_TYPES.

  • config – Hyper-parameter dictionary forwarded to the underlying classifier and trainer.

  • accelerator – Initial accelerator hint, 'cpu' or 'cuda'. Re-evaluated at sortie time by spike().

property clf_object: type

Classifier class associated with this unit’s type.

Returns:

The class (not an instance) registered under self.type in SUPPORTED_TYPES.

property id: str

Public unit ID combining the model type and tail.

Returns:

"{type}_{tail}" formatted identifier.

spike(accelerator: str, available_devices: list = None)

Verify that the unit is ready to launch and resolve its accelerator.

Performs three checks in order:

  1. The model type is supported by Ageas.

  2. CPU-only models are pinned to the CPU accelerator.

  3. CUDA models receive a valid device list, or fall back to CPU with a warning.

Parameters:

  • accelerator – Requested accelerator ('cpu' or 'cuda').

  • available_devices – Optional list of GPU device indices that the deck has reserved for this unit.

Returns:

The classifier class to instantiate, or None if the unit cannot be launched (unsupported type, unknown accelerator).