I started thinking a bit more about what we’re doing when we use something like posteriordb (from Stan) or Inference Gym (from TensorFlow Probability) to evaluate a sampler. posteriordb gives you Stan programs and 10,000 reference draws from their posterior. … Continue reading …