metropolis-hastings (MH) is a commonly-used mcmc algorithm, but it can be intractable on large datasets due to requiring computations over the whole dataset. In this paper, we study \emph{minibatch MH} methods, w