This project implements a complete Simulation-Based Inference (SBI) pipeline for complex ecological simulations. By using deep neural density estimators (e.g., Masked Autoregressive Flow), we recover the underlying parameters of an agent-based model from observed biodiversity data — without needing an explicit likelihood function.
Developed as part of an undergraduate capstone project (ICS 496 at University of Hawai‘i at Mānoa), this repository showcases an end-to-end example of Likelihood-Free Inference (LFI) applied to real-world ecological research.
Documentation Site: https://role-model.github.io/likelihood-free-inference/
- Agent-based simulation model for biodiversity
- Neural posterior estimation using PyTorch and the
sbilibrary - GPU acceleration with CUDA
- Full inference workflow and result visualization
- Jupyter notebooks for testing and experiments
git clone https://github.com/your-org/likelihood-free-inference
cd likelihood-free-inference
python3 -m venv .venv
source .venv/bin/activate
pip install -e .Project Sponsor: Andrew Rominger, PhD
Authors: Micah Tilton & Frances Uy