ytm.guided.backends.cuda.CUDADevice

class ytm.guided.backends.cuda.CUDADevice(config: ConfigT, device_config: DeviceConfig, state: dict | None = None)

Bases: CUDADevice[GuidedFitBuffers, TMConfig]

fit_epoch(X: ndarray, Y: ndarray, clause_drop_p: float, **kwargs)

Fit one epoch.

fit_sample(e: int) → float

Fit one sample e.

get_packed_clauses() → PackedClauses

Return packed clauses.

get_state_dict() → dict

Get state dictionary for saving the model.

infer(X: ndarray, force_repack: bool = False) → ndarray

Inference. Calculate class sums for X.

load_state_dict(state: dict) → None

Written in place, so anything already pointing at these arrays stays valid.

set_feat_drop(fids: int | list[int], value: int) → None

Set a feature to be ignored.

transform(X: ndarray, force_repack: bool = False) → ndarray

Calculate clause activations.

transform_patchwise(X: ndarray, force_repack: bool = False) → ndarray

Patchwise Clause activations.

wac(X: ndarray, target_classes: ndarray, polarity: int, force_repack: bool = False) → ndarray

Calculate WAC score.

wic(class_id: int, polarity: int, pw_th: float = 0.0, force_repack: bool = False) → ndarray

Calculate WIC score.