Performance of MCMC in Different Scenario

Abstract

Markov chain Monte Carlo (MCMC) is a method to obtain a sequence of random samples converging to be distributed to the target probability distribution which is hard to directly sample for. This capstone project explores an important and known version of MCMC, which is Hamiltonian Monte Carlo (HMC). This project explores the performance of HMC in different imbalanced dataset scenarios and diagnose the performance of HMC by using trace plots and autocorrelation function plots. And also this project explains why HMC performs well in some imbalanced data scenarios.