sbiax 0.0.0

Fast, parallel and lightweight simulation-based inference in JAX.

sbiax

Fast, lightweight and parallel simulation-based inference.

sbiax is a lightweight library for simulation-based inference (SBI) with a fixed-grid of simulations.

The design puts the neural density estimator (NDE) models at the centre of the code, allowing for flexible combinations of different models.

---

[!WARNING] :building_construction: Note this repository is under construction, expect changes. :building_construction:

Design

A typical inference with SBI occurs with

• fitting a density estimator to a set of simulations and parameters ${\xi, \pi}$ that may be compressed to summary statistics,
• the measurement of a datavector $\hat{\xi}$,
• the sampling of a posterior $p(\pi|\hat{\xi})$ conditioned on the measurement $\hat{\xi}$.

sbiax is designed to perform such an inference.

---

Usage

Install via

pip install sbiax

and have a look at examples.