Empirical verification of phase transitions in the Langevin algorithm for high dimensional Bayesian inference in the spiked 2 + 3-spin model

This is a mini repo for reproducing some of the main results in Mannelli et al. (2020). This is our course final project for AP229.

• run_langevin_gpu.py is the main script for running the Langevin algorithm.
• analysis.ipynb is where the figures are generated.
• compute_hamiltonian.py is an auxiliary script for computing the Hamiltonian for Langevin trajectories after they have been computed (for compute reasons we don't explicitly compute the Hamiltonian during Langevin sampling)