SBI-Comparison

This repository provides implementations of various Simulation-Based Inference (SBI) methods for comparative experiments. The included methods are:

• BayesFlow (Amortized Bayesian workflows)
• Flow Matching (Flow-based inference)
• Sequential Neural Likelihood (SNL) (Implemented in the sbi library)

Each method contains two experimental setups:

• gaussian: Gaussian likelihood experiments.
• disease: Disease modeling experiments.

Installation

Ensure you have Python 3.11.10 installed. To set up the environment, run:

git clone https://github.com/xudongwu-0/SBI-comparison.git
cd SBI-comparison

Fully Interactive Jupyter Notebooks

All experiments in this repository are designed to be fully interactive and are provided as Jupyter notebooks (.ipynb). Simply open any notebook, run the code, and modify parameters as needed to explore different settings in real-time.

Running the Notebooks

To launch the Jupyter environment, navigate to the repository and run:

pip install jupyter
jupyter notebook

Then, open any .ipynb file from the corresponding method folder (BayesFlow/, AFM/, or SNL/). These notebooks allow you to:

• Adjust prior settings, likelihood choices, and inference hyperparameters.
• Run experiments interactively without modifying core scripts.
• Visualize posterior distributions and calibration results in real-time.

This structure ensures that every experiment is reproducible and customizable with minimal effort.

Package Requirements

The experiments were conducted using:

• Python: Version 3.11.10.
• BayesFlow: BayesFlow Version 1.1.6, implemented in the BayesFlow framework for amortized Bayesian workflows.
• Flow Matching: BayesFlow Version 2.0.0, applied for flow matching experiments.
• Sequential Neural Likelihood (SNL): SBI Version 0.23.2, implemented in the sbi (Simulation-Based Inference) library.