TransferLab Training: Simulation-Based Inference

Welcome to the TransferLab Training: Simulation-Based Inference (SBI). This is the readme for participants of the training.

Simulation-based inference is a statistical method used to enable Bayesian inference when the likelihood function of a complex model is computationally intractable or unknown. It achieves this by using neural network-based density estimation methods to approximate a specific part of Bayes' theorem using data simulated from the model.

This training provides participants with a thorough understanding of the fundamental principles and methods of SBI, including when and why to use these methods instead of traditional likelihood-based inference techniques. The training also offers hands-on experience in applying SBI.

Schematic overview of the Simulation-based Inference workflow (Jan Boelts, 2023).

Training Agenda

You can find the agenda for the training in the file AGENDA. Everything is already set up, so you can either follow the trainer's presentation or explore the notebooks and source code on your own.

Running the Notebooks

Participants will have access to a Jupyter Hub during the training to run the notebooks. However, you might also want to pull the repository and execute it locally. Below are the steps for running the content:

Viewing the Notebooks

1. To simply view the rendered notebooks, open html/index.html in your browser.

Executing the Notebooks Locally

Without Docker

1. Create a conda environment with Python 3.11:

   conda create -n tfl_training_sbi python=3.11

2. Install the dependencies and the package:

   bash build_scripts/install_presentation_requirements.sh
   pip install -e .

3. Adapt the data path in config.yml to point to the correct location on your machine.

With Docker

1. Build the Docker image locally:

   docker build -t tfl_training_sbi .

2. Start the container:

   docker run -it -p 8888:8888 tfl_training_sbi jupyter notebook

Building Documentation

To create source code documentation, run:

bash build_scripts/build_docs.sh

Then, open docs/build/html/index.html in your browser. This will also rebuild the Jupyter Book-based notebook documentation originally found in the html directory.

Note: There is some non-trivial logic in the entrypoint that may collide with mounting volumes directly inside /home/jovyan/tfl_training_sbi. If you want to mount volumes there, the easiest way is to override the entrypoint or to mount somewhere else and create a symbolic link. For details, see the Dockerfile and entrypoint.sh.

---

Feel free to reach out if you have any questions or need assistance during the training. Enjoy your learning experience!

Collaborators

We would like to thank the The Machine Learning ⇌ Science Collaboratory and the Mackelab, Machine Learning in Science for their collaboration on this training. They shared materials of previous SBI workshops with us and helped with the initial design of the new training. Both are part of the Cluster of Excellence - Machine Learning for Science at the University of Tübingen.

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.