snowline 0.6.3

Fit and compare complex models quickly. Laplace Approximation, Variational Bayes, Importance Sampling.

snowline

Fit and compare models very quickly. MCMC-free.

About

Posterior distributions and corner plots without MCMC? No dealing with convergence criteria?

Yes!

Tailored for low-dimensional (d<10) problems with a single mode, this package automatically finds the best fit and uses the local covariance matrix as a Laplace Approximation. Then Importance Sampling and Variational Bayes come in to improve from a single-gaussian approximation to more complex shapes. This allows sampling efficiently in some problems, and provides a estimate for the marginal likelihood.

This package is built on top the excellent (i)minuit and pypmc packages.

You can help by testing snowline and reporting issues. Code contributions are welcome. See the Contributing page.

Features

• Pythonic. pip installable.

• Easy to program for: Sanity checks with meaningful errors

• Fast

• MPI support

Usage

Read the full documentation at:

https://johannesbuchner.github.io/snowline/

Licence

GPLv3 (see LICENCE file). If you require another license, please contact me.

Icon made by Vecteezy.

Other projects

See also:

• UltraNest: https://johannesbuchner.github.io/UltraNest/

• autoemcee: https://johannesbuchner.github.io/autoemcee/

Release Notes

0.4.0 (2020-03-07)

• Improve robustness to poor Laplace approximations

0.3.0 (2020-05-07)

• Numerical robustness

0.2.0 (2020-04-21)

• Robustness improvements

• Packaging and testing improvements

0.1.0 (2020-03-07)

• First version