Nested ratio estimation and inhomogeneous poisson point process sample caching for simulator efficient marginal posterior estimation.
Project description
Check out the quickstart notebook -->
Disclaimer: swyft is research software under heavy development and still in its alpha phase. There are many rough edges, and things might break. However, the core algorithms work, and we use swyft in production for research papers. If you encounter problems, please contact the authors or submit a bug report.
SWYFT
Neural nested marginal posterior estimation
Cursed by the dimensionality of your nuisance space? Wasted by Markov chains that reject your simulations? Exhausted from messing with simplistic models, because your inference algorithm cannot handle the truth? Try swyft for some pain relief.
A simple example is avaliable on google colab.
Our repository applying swyft to benchmarks and example inference problems is available at tmnre.
Installation
After installing pytorch, please run the command:
pip install swyft
Relevant Tools
swyft exists in an ecosystem of posterior estimators. The project sbi is particularly relevant as it is a collection of likelihood-free / simulator-based methods.
Citing
If you use swyft in scientific publications, please cite:
Truncated Marginal Neural Ratio Estimation. Benjamin Kurt Miller, Alex Cole, Patrick Forré, Gilles Louppe, Christoph Weniger. https://arxiv.org/abs/2107.01214
Simulation-efficient marginal posterior estimation with swyft: stop wasting your precious time. Benjamin Kurt Miller, Alex Cole, Gilles Louppe, Christoph Weniger. https://arxiv.org/abs/2011.13951
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.