I increased the test coverage of variational_optimization.py from 79% -> 90% by adding three tests using Gaussian Mixtures and removing some dead code. The results and the comparison to MATLAB can be found below.
Furthermore, after speaking to @lacerbi I removed the CMA-ES optimization as it does not work well with variational_optimization. Then, I fixed a bug resulting from vp.sigma being 1D in variational_optimization.py. Finally, I added a setup.py so that we are able to use pyvbmc in jupiter notebooks without having to setup the final installation procedure and replaced some generic exceptions by specific ones.
I increased the test coverage of
variational_optimization.py
from 79% -> 90% by adding three tests using Gaussian Mixtures and removing some dead code. The results and the comparison to MATLAB can be found below.Furthermore, after speaking to @lacerbi I removed the CMA-ES optimization as it does not work well with variational_optimization. Then, I fixed a bug resulting from
vp.sigma
being 1D invariational_optimization.py
. Finally, I added a setup.py so that we are able to use pyvbmc in jupiter notebooks without having to setup the final installation procedure and replaced some generic exceptions by specific ones.