ytm._core.initializer.SampleTAInit

class ytm._core.initializer.SampleTAInit(X: ndarray, Y: ndarray | None = None, margin: int = 0, p: float = 0.0, include_position: bool = False)

Bases: TAInit

Start each clause as a copy of a training sample.

Each clause picks a sample (and, with convolution, a random patch of it) and includes exactly the literals that are true for it. All other literals start just below include.

Parameters:

Xarray-like of shape (N, …)

Samples to copy from, in the same format as fit takes.

Yarray-like, default=None

Labels for X, in the same format as fit takes. Without Y, samples are picked at random. With Y, a clause with positive weight for a class copies a sample of that class, and a clause with negative weight copies a sample not of that class.

marginint, default=0

Spread of the initial states. Included literals start uniformly in [include_state, include_state + margin], excluded ones in [include_state - 1 - margin, include_state - 1].

pfloat, default=0.0

Probability, in [0, 1), of leaving out each copied literal.

include_positionbool, default=False

Also include the position literals of the copied patch.