mangoes.weighting module¶

Weighting transformations to apply to the co-occurrence count matrix.

This module provides transformations that can be used in the transformations parameter of the mangoes.create_representation() function to create an Embeddings from a CountBasedRepresentation.

Examples¶

>>> import mangoes.base
>>> ppmi = mangoes.weighting.PPMI(alpha=1)
>>> embeddings = mangoes.base.create_representation(cc, transformations=ppmi)

See Also¶

mangoes.create_representation() mangoes.Transformation

class mangoes.weighting.JointProbabilities¶

Bases: mangoes.base.Transformation

Defines a transformation that replaces counts with joint probabilities computed from these counts

Attributes

params

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the matrix [P(wi,cj)].

class mangoes.weighting.ConditionalProbabilities¶

Bases: mangoes.base.Transformation

Defines a transformation that replaces counts with joint probabilities computed from these counts

Attributes

params

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the matrix [P(cj|wi)].

class mangoes.weighting.ProbabilitiesRatio(alpha=1)¶

Bases: mangoes.base.Transformation

Defines a transformation that replaces counts with joint probabilities computed from these counts

Attributes

alpha: int, optional: positive number (default=1); “smoothing” parameter for the computation of the context probability distributions: P(c_j) = (#c_j)**alpha / sum((#c_k)**alpha)

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the matrix [P(cj|wi) / P(cj)].

property alpha¶

class mangoes.weighting.PMI(alpha=1)¶

Bases: mangoes.base.Transformation

Defines a PMI (Pointwise Mutual Information) transformation

Attributes

alpha: int, optional: positive number (default=1); “smoothing” parameter for the computation of the context probability distributions: P(c_j) = (#c_j)**alpha / sum((#c_k)**alpha)

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the ‘pmi’ matrix [log(P(cj|wi)/P(cj)) or 0 if not defined].

property alpha¶

class mangoes.weighting.PPMI(alpha=1)¶

Bases: mangoes.weighting.PMI

Defines a PPMI (Positive PMI) transformation.

Returns

matrix-like object

Attributes

alpha: int, optional: positive number (default=1); “smoothing” parameter for the computation of the context probability distributions: P(c_j) = (#c_j)**alpha / sum((#c_k)**alpha)

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the ‘ppmi’ matrix [max(log(P(cj|wi)/P(cj)) or 0 if not defined, 0)].

class mangoes.weighting.ShiftedPPMI(alpha=1, shift=1)¶

Bases: mangoes.weighting.PMI

Defines a Shifted PPMI (Positive PMI) transformation

From matrix = [#(wi,cj)], compute the matrix [max(log(P(cj|wi)/P(cj))-log(shift) or 0 if not defined, 0)]

Returns

matrix-like object

Attributes

alpha: int, optional: positive number (default=1); “smoothing” parameter for the computation of the context probability distributions: P(c_j) = (#c_j)**alpha / sum((#c_k)**alpha)
shift: int, optional: number superior or equal to 1 (default=1); The result matrix will be shifted of log(shift)

Methods

__call__(matrix)

From matrix = [#(wi,cj)], compute the matrix [max(log(P(cj|wi)/P(cj))-log(shift) or 0 if not defined, 0)].

property shift¶

class mangoes.weighting.TFIDF¶

Bases: mangoes.base.Transformation

Defines a TF-IDF (term frequency-inverse document frequency) transformation

Attributes

params

Methods

__call__(matrix)

From matrix = [#(wi,cj)], computes a tfidf matrix [tf(wi,cj) * idf(wi)]