sbi: simulation-based inference Save

sbi: simulation-based inference

Getting Started | Documentation

sbi is a PyTorch package for simulation-based inference. Simulation-based inference is the process of finding parameters of a simulator from observations.

sbi takes a Bayesian approach and returns a full posterior distribution over the parameters of the simulator, conditional on the observations. The package implements a variety of inference algorithms, including amortized and sequential methods. Amortized methods return a posterior that can be applied to many different observations without retraining; sequential methods focus the inference on one particular observation to be more simulation-efficient. See below for an overview of implemented methods.

sbi offers a simple interface for posterior inference in a few lines of code

from sbi.inference import SNPE
# import your simulator, define your prior over the parameters
# sample parameters theta and observations x
inference = SNPE(prior=prior)
_ = inference.append_simulations(theta, x).train()
posterior = inference.build_posterior()

Installation

sbi requires Python 3.8 or higher. A GPU is not required, but can lead to speed-up in some cases. We recommend to use a conda virtual environment (Miniconda installation instructions). If conda is installed on the system, an environment for installing sbi can be created as follows:

# Create an environment for sbi (indicate Python 3.8 or higher); activate it
$ conda create -n sbi_env python=3.10 && conda activate sbi_env

Independent of whether you are using conda or not, sbi can be installed using pip:

pip install sbi

To test the installation, drop into a python prompt and run

from sbi.examples.minimal import simple
posterior = simple()
print(posterior)

Tutorials

For first time users, you can now head over to the turorials and get going with Getting Started.

Inference Algorithms

The following inference algorithms are currently available. You can find instructions on how to run each of these methods here.

Neural Posterior Estimation: amortized (NPE) and sequential (SNPE)

• SNPE_A (including amortized single-round NPE) from Papamakarios G and Murray I Fast ε-free Inference of Simulation Models with Bayesian Conditional Density Estimation (NeurIPS 2016).

• SNPE_C or APT from Greenberg D, Nonnenmacher M, and Macke J Automatic Posterior Transformation for likelihood-free inference (ICML 2019).

• TSNPE from Deistler M, Goncalves P, and Macke J Truncated proposals for scalable and hassle-free simulation-based inference (NeurIPS 2022).

Neural Likelihood Estimation: amortized (NLE) and sequential (SNLE)

• SNLE_A or just SNL from Papamakarios G, Sterrat DC and Murray I Sequential Neural Likelihood (AISTATS 2019).

Neural Ratio Estimation: amortized (NRE) and sequential (SNRE)

• (S)NRE_A or AALR from Hermans J, Begy V, and Louppe G. Likelihood-free Inference with Amortized Approximate Likelihood Ratios (ICML 2020).

• (S)NRE_B or SRE from Durkan C, Murray I, and Papamakarios G. On Contrastive Learning for Likelihood-free Inference (ICML 2020).

• BNRE from Delaunoy A, Hermans J, Rozet F, Wehenkel A, and Louppe G. Towards Reliable Simulation-Based Inference with Balanced Neural Ratio Estimation (NeurIPS 2022).

• (S)NRE_C or NRE-C from Miller BK, Weniger C, Forré P. Contrastive Neural Ratio Estimation (NeurIPS 2022).

Neural Variational Inference, amortized (NVI) and sequential (SNVI)

• SNVI from Glöckler M, Deistler M, Macke J, Variational methods for simulation-based inference (ICLR 2022).

Mixed Neural Likelihood Estimation (MNLE)

• MNLE from Boelts J, Lueckmann JM, Gao R, Macke J, Flexible and efficient simulation-based inference for models of decision-making (eLife 2022).

Feedback and Contributions

We welcome any feedback on how sbi is working for your inference problems (see Discussions) and are happy to receive bug reports, pull requests and other feedback (see contribute). We wish to maintain a positive community, please read our Code of Conduct.

Acknowledgements

sbi is the successor (using PyTorch) of the delfi package. It was started as a fork of Conor M. Durkan's lfi. sbi runs as a community project. See also credits.

Support

sbi has been supported by the German Federal Ministry of Education and Research (BMBF) through project ADIMEM (FKZ 01IS18052 A-D), project SiMaLeSAM (FKZ 01IS21055A) and the Tübingen AI Center (FKZ 01IS18039A).

License

Affero General Public License v3 (AGPLv3)

Citation

If you use sbi consider citing the sbi software paper, in addition to the original research articles describing the specific sbi-algorithm(s) you are using.

@article{tejero-cantero2020sbi,
  doi = {10.21105/joss.02505},
  url = {https://doi.org/10.21105/joss.02505},
  year = {2020},
  publisher = {The Open Journal},
  volume = {5},
  number = {52},
  pages = {2505},
  author = {Alvaro Tejero-Cantero and Jan Boelts and Michael Deistler and Jan-Matthis Lueckmann and Conor Durkan and Pedro J. Gonçalves and David S. Greenberg and Jakob H. Macke},
  title = {sbi: A toolkit for simulation-based inference},
  journal = {Journal of Open Source Software}
}

The above citation refers to the original version of the sbi project and has a persistent DOI. Additionally, new releases of sbi are citable via Zenodo, where we create a new DOI for every release.