Analysis of the paper A Non-Asymptotic Analysis for Stein Variational Gradient Descent, and detailed summary of the theoretical results of SVGD for the Bayesian Machine Learning course at the MVA master at ENS Paris-Saclay.
One might want to read the report first.
We also provide an implementation of the SVGD algorithm using Numpy.
Install the required packages by running:
pip install -r requirements.txt
if you don't.
Then, run the code by running:
python main.py nb_iterations
it will run both experiments detailed in the report
and save all the figures in a exp[1-2]
folder.