Parallel Affine Transformation Tuning of MCMC

This directory contains all numerical experiments (i.e. data, source code and plots) belonging to the paper

Philip Schär, Michael Habeck, Daniel Rudolf
Parallel Affine Transformation Tuning of Markov Chain Monte Carlo
Proceedings of the 41st International Conference on Machine Learning,
PMLR 235, pp. 43571-43607, 2024

available in the conference proceedings. The directory code contains the entire code base used to run the experiments (consisting of various python modules) as well as the experiments themselves (each consisting of a Jupyter notebook). See the readme therein for details.

The directory data contains the data sets used in the experiments. References for where we obtained these data sets are contained in the experiment notebooks based on them.

Finally, the directory plots contains all plots presented in the paper as well as some additional ones that were also generated by the experiment notebooks.

Recommended Setup for Reproduction

Note: The following guide presupposes working on a relatively up-to-date linux system.

Open a terminal in the directory you want to clone this repo into. Proceed by sequentially executing the following commands in it.

1. Install pyenv:
   curl -fsSL https://pyenv.run | bash
2. Use pyenv to safely install python 3.12.11 alongside whatever version the system normally uses:
   pyenv install 3.12.11
3. Create a venv for the PATT experiments, based on the local python 3.12 installation:
   ~/.pyenv/versions/3.12.11/bin/python -m venv ~/.venvs/patt_experiments
4. Clone this repo and move into its code/ directory:
   git clone git@github.com:microscopic-image-analysis/patt_mcmc.git patt_mcmc
cd patt_mcmc/code/
5. Install the python packages required to run the experiments with the venv's pip:
   ~/.venvs/patt_experiments/bin/pip install -r requirements.txt
6. Append your .bashrc by a convenient shortcut for launching the venv's classic jupyter client, then restart the terminal to make the alias available:
   echo 'alias patt_experiments_nb=~/.venvs/patt_experiments/bin/jupyter-nbclassic' >> ~/.bashrc
exec $SHELL

After completing the above steps once, you should be able to navigate and run the experiments through jupyter, specifically by executing patt_experiments_nb
in the directory patt_mcmc/code/ to launch the patt_experiments venv's jupyter client.