A python package for working with Gaussian Mixture Models. You can either generate your own data and fit models to help understand how the models work, or you can work directly with another non-synthetic data source.
git clone https://github.com/blazickjp/mixtures.git
cd mixtures
pip install .
Define a mixture model and generate sample data according to the model. We can then fit a mixture model to the data and compare our fitted model to actuals.
import mixtures
import numpy as np
gmm = mixtures.FiniteGMM(k = 4, mu = [0,4,8,16], sigma = [1,1,2,3], phi = [.2,.2,.2,.4])
gmm.data_gen(1000)
gmm.plot_data(alpha = .5, bins = 50)
gmm.gibbs(iters=500, burnin=50)
gmm.plot_results(alpha = .5, bins=50)
We can also generate and fit multitvariate data by simply providing multivariate parameters to the model. The FiniteGMM class will detect the shape of the inputs and set self.multivariate = True. This tells the class to take different actions on the inputs and uses a collapsed gibbs sampler for fitting the model.
# Define parameters for K=3 Mixture of Multivariate Gaussians
phi = [.3, .5, .2]
mu = np.array([[13,5], [0,-2], [-14,3]])
cov_1 = np.array([[2.0, 0.3], [0.3, 0.5]])
cov_2 = np.array([[3.0,.4], [.4,3.0]])
cov_3 = np.array([[1.7,-.7], [-.7,1.7]])
cov = np.stack((cov_1, cov_2, cov_3), axis = 0)
gmm = mixtures.FiniteGMM(3, mu, cov, phi)Now we can use the same functionality as in the univariate case.
gmm.data_gen(1000)
gmm.plot_data(alpha = .5)
gmm.gibbs(iters=20)
gmm.plot_results(alpha = .5)
The API for working with Infitine Mixture Models is exactly the same except we call InfitineGMM instead of FiniteGMM. The Infinite GMM model will fit both multivariate and univarite data. The InfiniteGMM class inherits from the FiniteGMM class so you get the sample plotting features as in the finite class.
gmm = mixtures.InfiniteGMM(4, [0,4,8,16], [1,1,2,3], [.2,.2,.2,.4])
gmm.data_gen(1000)
gmm.gibbs(initial_k=3,iters=50)
gmm.plot_results()
