Conjugate Bayesian Models
Last update: June 2019.
Lightweight Python library implementing a few conjugate Bayesian models. For details on the derivations see [1].
pip3 install conjugate-bayes
We support the following:
To fit distribution models
- Beta-Bernoulli
- Gamma-Poisson
- Normal-Inverse-Gamma
To fit regression models
- Linear regression with Normal Inverse-Gamma prior
- Linear regression with Zellner's g-prior
Future work
- Dirichlet-Multinomial
- Normal-Inverse-Wishart
Usage
Below we show an example fitting a simple Bayesian linear regression with unknown beta and unknown variance.
import numpy as np
from conjugate_bayes.models import NIGLinearRegression
model = NIGLinearRegression(mu=np.zeros(2), v=100*np.eye(2), a=0.5, b=0.5)
model.fit(x_tr, y_tr)
# these are scipy distribution objects
sigma2 = model.get_marginal_sigma2()
beta = model.get_conditional_beta(sigma2=sigma2.mean())The above example results in the following prediction intervals.
For further details the examples/ folder.
References
[1] P. D. Hoff, A First Course in Bayesian Statistical Methods (New York: Springer-Verlag, 2009).
License
This library is available under the MIT License.