pyrichlet 0.0.4

A package for density estimation and clustering using infinite Gaussian mixtures with stick-breaking weighting structures

Project description

Pyrichlet is a package for doing data analysis via density estimation and clustering using Gaussian mixtures of several weighting structures.

Installation

With pip:

pip install pyrichlet

For a specific version:

pip install pyrichlet==0.0.3

Usage

This is a quick guide. For a more detailed usage see https://pyrichlet.readthedocs.io/en/latest/index.html.

The weighting structure models that this package implements are

• DirichletDistribution
• DirichletProcess
• PitmanYorProcess
• GeometricProcess
• BetaInBetaProcess
• BetaInDirichletProcess
• BetaBernoulliProcess
• BetaBinomialProcess

They can be imported and initialized as

from pyrichlet import weight_models

wm = weight_models.DirichletProcess()
wm.fit([0, 0, 1])
wm.random(10)
wm.random_assignment(100)
wm.reset()
wm.random(10)
wm.random_assignment(100)

For each weighting structure there is an associated Gaussian mixture model, formerly

• DirichletDistributionMixture
• DirichletProcessMixture
• PitmanYorMixture
• GeometricProcessMixture
• BetaInBetaMixture
• BetaInDirichletMixture
• BetaBernoulliMixture
• BetaBinomialMixture

The mixture models can fit array or dataframe data for density estimation

from pyrichlet import mixture_models

mm = mixture_models.DirichletProcessMixture()
y = [1, 2, 3, 4]
mm.fit_gibbs(y)
x = 2.5
f_x = mm.gibbs_eap_density(x)

mm.fit_variational(y, n_groups=2)
f_x = mm.var_eap_density(x)

or for cluster estimation

mm.var_map_cluster()
mm.gibbs_map_cluster()
mm.gibbs_eap_spectral_consensus_cluster()