src.lscd.permutation

Contents

`src.lscd.permutation`#

Module Contents#

Classes#

class src.lscd.permutation.Permutation#

Bases: src.lscd.model.GradedLSCDModel

wic: src.wic.ContextualEmbedder#

n_perms: int#

whiten: bool#

k: int | None#

static compute_kernel_bias(vecs: numpy.typing.NDArray[numpy.float32], k: int | None = None) → tuple[numpy.typing.NDArray[numpy.float32], numpy.typing.NDArray[numpy.float32]]#: vecs = matrix (n x 768) with the sentence representations of your whole dataset (in the paper they use train, val and test sets)

static transform_and_normalize(vecs: numpy.typing.NDArray[numpy.float32], kernel: numpy.typing.NDArray[numpy.float32] | None = None, bias: numpy.typing.NDArray[numpy.float32] | None = None) → numpy.typing.NDArray[numpy.float32]#: Kernel and bias are W and -mu from previous function. They’re passed to this function when inputing vecs vecs = vectors we need to whiten.

static euclidean_dist(m0: numpy.typing.NDArray[numpy.float32], m1: numpy.typing.NDArray[numpy.float32]) → numpy.typing.NDArray[numpy.float32]#

get_n_rows(len_m0: int, len_m1: int) → int#

permute_indices(len_m0: int, len_m1: int) → tuple[list[int], list[int]]#

predict(lemma: src.lemma.Lemma) → float#

predict_all(lemmas: list[src.lemma.Lemma]) → list[float]#