Burn-in and Warmup
Early iterations reflect the arbitrary starting point rather than the stationary distribution and are discarded. Modern samplers also use this warmup phase to adapt tuning parameters (step size, mass matrix), so adapted iterations must never be kept as samples.
This Concept is waiting for its first lesson!
Early iterations reflect the arbitrary starting point rather than the stationary distribution and are discarded. Modern samplers also use this warmup phase to adapt tuning parameters (step size, mass matrix), so adapted iterations must never be kept as samples.
Are you a teacher? Sign in to start contributing.
Sign In