Histogram GMM

A Gaussian Mixture Model implementation that works with a histogram instead of using raw data points.

This implementation is useful for fitting data that has the form of a sum of gaussian curves, but it can also be used as a typical GMM implementation by computing a histogram of data points.

• Documentation: https://histgmm.readthedocs.io/en/latest/histgmm.html

Installation

As of now, this is not on PyPI, so the easiest way to install it is by cloning this repo and then installing it with pip:

git clone https://github.com/thomasvf/histgmm.git
cd histgmm
pip install .

Usage

The example below shows how HistogramGMM can be used to fit the GMM to a sum of gaussians. It's also available as a Jupyter notebook here.

import numpy as np

from histgmm import HistogramGMM
from histgmm.utils import gaussian_1d

# create data for the example
x = np.arange(0, 100, 1)
h = gaussian_1d(x, A=0.2, mu=10, var=9) + \
    gaussian_1d(x, A=1, mu=35, var=16) + \
    gaussian_1d(x, A=0.7, mu=46, var=25) + \
    gaussian_1d(x, A=1, mu=65, var=25)


# fit using 4 components
model = HistogramGMM(n_components=4, n_dimensions=1)
model.fit(x.reshape(-1, 1), h)

The fitted gaussians can be plotted over the original data using the function plot_1d_gaussian_fit, as shown below.

import matplotlib.pyplot as plt
from histgmm.visualization import plot_1d_gaussian_fit

fig, ax = plt.subplots(1, 2, figsize=(12, 4))
ax[0].plot(x, h, marker='.')
ax[0].set_title('Original Curve')

plot_1d_gaussian_fit(gmm=model, X=x.reshape(-1, 1), h=h, ax=ax[1])
ax[1].set_title('Fitted and Scaled Gaussians')

plt.show()

Model Formulation

See the PDF in here.

Credits

This package was created with Cookiecutter_ and the audreyr/cookiecutter-pypackage_ project template.

.. _Cookiecutter: https://github.com/audreyr/cookiecutter .. _audreyr/cookiecutter-pypackage: https://github.com/audreyr/cookiecutter-pypackage