Closed BradyPlanden closed 1 day ago
HMC (or NUTS is the varient always used in practice) is great. But for low numbers of parameters a simple method like Adaptive Covariance can be more reliable. Either way, loads of libraries give you NUTS and Adaptive Covariance (like PINTS ;) ) so I think getting both of these will be easy
The No-U-Turn Hamiltonian Monte Carlo (HMC) method provide a robust method to capture a posterior distributions with efficient sampling and adaptive step size selection. Capturing the parameter posterior distributions for $\varepsilon = (\hat{y} - y$). This issue surmises the following: