scipy.stats.

Mixture

class scipy.stats.Mixture(components, *, weights=None)

Representation of a mixture distribution.

A mixture distribution is the distribution of a random variable defined in the following way: first, a random variable is selected from components according to the probabilities given by weights, then the selected random variable is realized.

Parameters:

componentssequence of UnivariateDistribution

The underlying instances of UnivariateDistribution. All must have scalar shape parameters (if any); e.g., the mean must return a scalar.

weightssequence of floats, optional

The corresponding probabilities of selecting each random variable. Must be non-negative and sum to one. The default behavior is to weight all components equally.

Attributes:

componentssequence of UnivariateDistribution

The underlying instances of UnivariateDistribution.

weightsndarray

The corresponding probabilities of selecting each random variable.

Methods

support()	Support of the random variable
sample([shape, rng, method])	Random sample from the distribution.
moment([order, kind, method])	Raw, central, or standard moment of positive integer order.
mean(*[, method])	Mean, a measure of location
median(*[, method])	Median, a measure of location
mode(*[, method])	Mode, the most likely value
variance(*[, method])	Variance, a measure of dispersion
standard_deviation(*[, method])	Standard deviation, a measure of dispersion
skewness(*[, method])	Skewness, a measure of asymmetry
kurtosis(*[, method])	Kurtosis, a measure of tailedness
pdf(x, /, *[, method])	Probability density function
logpdf(x, /, *[, method])	Log of the probability density function
cdf(x[, y, method])	Cumulative distribution function
icdf(p, /, *[, method])	Inverse of the cumulative distribution function.
ccdf(x[, y, method])	Complementary cumulative distribution function
iccdf(p, /, *[, method])	Inverse complementary cumulative distribution function.
logcdf(x[, y, method])	Log of the cumulative distribution function
ilogcdf(p, /, *[, method])	Inverse of the logarithm of the cumulative distribution function.
logccdf(x[, y, method])	Log of the complementary cumulative distribution function
ilogccdf(p, /, *[, method])	Inverse of the log of the complementary cumulative distribution function.
entropy(*[, method])	Differential entropy

Notes

Methods median, mode, entropy, logentropy, icdf, iccdf, ilogcdf, and ilogccdf are currently implemented only when all components are continuous distributions.

The following abbreviations are used throughout the documentation.

• PDF: probability density function

• CDF: cumulative distribution function

• CCDF: complementary CDF

• entropy: differential entropy

• log-F: logarithm of F (e.g. log-CDF)

• inverse F: inverse function of F (e.g. inverse CDF)

Array API Standard Support

Mixture has support for Python Array API Standard compatible backends in addition to NumPy. The following combinations of backend and device (or other capability) are supported.

Library	CPU	GPU
NumPy	✅	n/a
CuPy	n/a	✅
PyTorch	⛔	⛔
JAX	⛔	⛔
Dask	⛔	n/a

See Support for the array API standard for more information.

References

[1]

Mixture distribution, Wikipedia, https://en.wikipedia.org/wiki/Mixture_distribution

[2]

Zero-inflated model, Wikipedia, https://en.wikipedia.org/wiki/Zero-inflated_model

Examples

Try it in your browser!

A mixture of normal distributions:

>>> import numpy as np
>>> from scipy import stats
>>> import matplotlib.pyplot as plt
>>> X1 = stats.Normal(mu=-2, sigma=1)
>>> X2 = stats.Normal(mu=2, sigma=1)
>>> mixture = stats.Mixture([X1, X2], weights=[0.4, 0.6])
>>> print(f'mean: {mixture.mean():.2f}, '
...       f'median: {mixture.median():.2f}, '
...       f'mode: {mixture.mode():.2f}')
mean: 0.40, median: 1.04, mode: 2.00
>>> x = np.linspace(-10, 10, 300)
>>> plt.plot(x, mixture.pdf(x))
>>> plt.title('PDF of normal distribution mixture')
>>> plt.show()

A zero-inflated Poisson model.

>>> Poisson = stats.make_distribution(stats.poisson)
>>> lmb = 10  # expected Poisson count
>>> pi = 0.4  # probability of extra zeros
>>> X = stats.Binomial(n=0, p=0)  # degenerate distribution; all zero
>>> Y = Poisson(mu=lmb)
>>> ZIP = stats.Mixture((X, Y), weights=(pi, 1-pi))
>>> true_mean = (1 - pi) * lmb  # see e.g. [2]
>>> print(np.allclose(ZIP.mean(), true_mean))
True
>>> true_variance = lmb*(1 - pi)*(1 + pi*lmb)  # see e.g. [2]
>>> print(np.allclose(ZIP.variance(), true_variance))
True

Go BackOpen In Tab