Mixture weights optimisation for Alpha-Divergence Variational Inference

This project can be used to reproduce the experiments presented in

Kamélia Daudel and Randal Douc. "Mixture weights optimisation for Alpha-Divergence Variational Inference". Accepted at Neurips 2021.

Parts of this implementation were modified from the code provided here and the code has been developed and tested with python3.6.

Reproducing the figures in the paper

To run the experiments, run:

python mainMixtureModel.py

To create and save the figures, run:

python plotResults.py

The figures can then be found in

.
├── results/

Choice of the Exploration step

Setting

main_on = True

in both mainMixtureModel.py and plotResults.py, will select the Exploration step described in Section 5 of the paper.

Setting

main_on = False

in both mainMixtureModel.py and plotResults.py, will select the Exploration step described in Appendix D.3.2 of the paper.

Parallelisation

Note that there is a possibility to parallelise the code via the joblib package.

To do so, in mainMixtureModel.py replace

#Parallel(nb_cores_used)(delayed(main_function)(i) for i in i_list) 
for i in i_list:
    main_function(i) 

by

Parallel(nb_cores_used)(delayed(main_function)(i) for i in i_list) 
#for i in i_list:
#    main_function(i) 

Feedback

Feedback is greatly appreciated. If you have any questions, feel me to send me an email.

All rights reserved.