Model Comparison In Approximate Bayesian Computation
mcabc is a Python package for performing model comparison in Approximate Bayesian Computation. It enables you to do Bayesian model comparison between models for which a likelihood function is not available. In contrast to previous approaches that are mainly based on variants of rejection sampling, this is a density estimation method that approximates the posterior over models in parametric form using a mixture-density network.
The implementation is part of my master thesis "Model Comparison in Approximate Bayesian Computation". It is work in progress and tailored to the examples presented in the thesis. If you are interested in applying it to a model comparison problem of yours, please let me know.
Clone the repository. Then, in the repository root folder, run
python setup.py install --user
to install all required packages. Alternatively, you install it via pip using
pip install git+https://github.com/janfb/mcabc.git --process-dependency-links
To test whether everything worked out fine, run
nosetests tests/
in the repository root folder.
You find a jupyter notebook in the examples folder presenting the methods on a tractable
example problem: Bayesian model comparison between a Poisson model and a negative binomial model.
More examples will follow soon.
The examples can be found on binder as well:
Please let me know if you have questions or suggestions.