GaussianMixture

Mixture of Gaussians fitted by expectation-maximization.

A smooth reference for one variable (NormalScore(reference=...)) or for several, e.g. two mineral associations that one Gaussian cannot represent. Rows may miss variables (NaN): each weighs the components by its present entries. The components start from a k-means++ clustering drawn from seed, so a seed gives one fit.

Parameters:

Name	Type	Description	Default
components	int	Number of Gaussians; by default the count from 1 to max_components with the lowest BIC.	required
max_components	int		6
seed	int		0

Attributes:

Name	Type	Description
proportions_	(ndarray, shape(k))
means_	(ndarray, shape(k, d))
covariances_	(ndarray, shape(k, d, d))
bic_	dict	Bayesian information criterion by number of components fitted.

fit method descriptor

fit(data, *, weights=None)

Fits the mixture.

Parameters:

Name	Type	Description	Default
data	(array_like, shape(n) or (n, d))	One variable or d; NaN marks a missing entry.	required
weights	array_like	Declustering weights of the rows.	None

from_json staticmethod

from_json(text)

Reads to_json output; raises InvalidInput on another class's JSON or a newer format.

pdf method descriptor

pdf(data)

Mixture density at each complete row of data.

predict method descriptor

predict(data)

Most probable component of each row given its present entries.

sample method descriptor

sample(n, *, seed=0)

n rows drawn from the mixture, shape (n, d); the same seed gives the same rows.

to_json method descriptor

to_json()

JSON of the parameters and, once fitted, the fitted state.

used in

• reference distributions