/sbi_intro

Tutorial on Simulation Based Inference

Primary LanguageJupyter Notebook

Introduction to SBI (and Scaling!)

Tutorial on Simulation Based Inference

Material of this tutorial was largely built using the examples from the repositories listed below:

https://github.com/sbi-dev/sbi/tree/main/tutorials

https://github.com/mlcolab/sbi-workshop/tree/main/slides

https://github.com/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers

Useful references:

The frontier of simulation-based inference, Kyle Cranmer, Johann Brehmer, and Gilles Louppe, PNAS 117 (48) 30055-30062 https://doi.org/10.1073/pnas.191278911

Fast ε-free Inference of Simulation Models with Bayesian Conditional Density Estimation, George Papamakarios, Iain Murray, NeurIPS 2016, https://proceedings.neurips.cc/paper_files/paper/2016/file/6aca97005c68f1206823815f66102863-Paper.pdf

Sequential neural likelihood: Fast likelihood-free inference with autoregressive flows, George Papamakarios, David C. Sterratt, Iain Murray, http://proceedings.mlr.press/v89/papamakarios19a/papamakarios19a.pdf

Likelihood-free MCMC with Amortized Approximate Likelihood Ratios, Joeri Hermans, Volodimir Begy, Gilles Louppe Proceedings of the 37th International Conference on Machine Learning http://proceedings.mlr.press/v119/hermans20a.html

Working with ray cluster

An example of how to submit jobs with ray backend for distributed training is available withing the folder ray_cluster

Use source ray_env/activate.sh to activate the corresponding environment. Then use the batch script run_ray_on_slurm.sbatch in order to submit an example script ray_joblib.py