sde-importance-sampling

This Python repository contains the implementation of importance sampling strategies for metastable diffusion processes. We build a framework to estimate path functionals of stochastic process following an overdamped Langevin equation.

Contains

• Monte Carlo sampling
• Adapted metadynamics algorithm for path dependent problems
• Finite differences solver of the associated HJB equation
• Importance sampling given a control (null, random, meta, hjb)
• Related stochastic optimization control (SOC) problem with different control representations (ansatz type and neural networks) and different initializations (null, random, meta, hjb)

Install

1. clone the repo

git clone git@github.com:riberaborrell/sde_importance_sampling.git

2. set python version

pyenv local 3.9.7

3. create virtual environment and install required packages

make venv

4. activate venv

source venv/bin/activate

Developement

in step 3) also install developement packages

make develop