Population Monte Carlo with Normalizing Flow

This is an implementation of the methodology in "Population Monte Carlo with Normalizing Flow". The arXiv version is available at https://arxiv.org/pdf/2312.03857.pdf

   

To reproduce the results in the paper: choose 'example' and 'sigma_prop' for running different experiments with different settings, refer to our paper for the citations of the baseline algorithms

1. PMC: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'Resample' --resampling 'global' --weighting 'Standard'

2. GR-PMC: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'Resample' --resampling 'global' --weighting 'DM'

3. LR-PMC: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'Resample' --resampling 'local' --weighting 'DM'

4. SL-PMC: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'Langevin' --resampling 'local' --weighting 'DM'

5. O-PMC: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'Newton' --resampling 'local' --weighting 'DM'

6. GRAMIS: python run_trials.py --num_trials 100 --example 'GMM'/'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'GRAMIS' --resampling 'local' --weighting 'DM'

7. HAIS (GMM): python run_trials.py --num_trials 100 --example 'GMM' --sigma_prop 1.0/2.0/3.0 --adaptation 'HMC' --resampling 'local' --weighting 'DM' --L_hmc 50 --eps_hmc 0.005

8. HAIS (Logistic): python run_trials.py --num_trials 100 --example 'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'HMC' --resampling 'local' --weighting 'DM' --L_hmc 50 --eps_hmc 0.05

9. VAPIS (GMM): python run_trials.py --num_trials 100 --example 'GMM' --sigma_prop 1.0/2.0/3.0 --adaptation 'VAPIS' --resampling 'local' --weighting 'DM' --lr_vi 0.25

10. VAPIS (Logistic): python run_trials.py --num_trials 100 --example 'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'VAPIS' --resampling 'local' --weighting 'DM' --lr_vi 0.5

11. NF-PMC (GMM) (ours): python run_trials.py --num_trials 100 --example 'GMM' --sigma_prop 1.0/2.0/3.0 --adaptation 'NF' --resampling 'local' --weighting 'DM' --lr_nf 0.005 --step_nf 25 --gamma 0.1

12. NF-PMC (Logistic) (ours): python run_trials.py --num_trials 100 --example 'Logistic' --sigma_prop 1.0/2.0/3.0 --adaptation 'NF' --resampling 'local' --weighting 'DM' --lr_nf 0.05 --step_nf 25 --gamma 0.1

 

Then the results will be saved in the ./results.

 

Calculation of the performance metrics:

Run python calculate_error.py

   

Requirements

See requirements.txt

Cite

Please cite our paper if you use this code in your work:

@article{pal2023, 
author={S. Pal and A. Valkanas and M. Coates}, 
journal={IEEE Sig. Process. Lett.}, 
title={{Population Monte Carlo} with Normalizing Flow},
month={},
year={2023}}