Improving metropolis-hastings code

[fipelle]

@marouane

if there is any minimum example for moving the coordinates one at a time

I am not sure about examples in Julia. However, pseudo-codes are widely available. If you are into adaptive Metropolis algorithms, you could take a look at the adaptive single component version of Haario et al. (2005). I have written it in Julia once, but the replication code is not on Github.

@opera_malenky

an acceptance rate of about 23.4% is often (though certainly not always) optimal.

True for most standard MHs. However, if @marouane is going to use a single component algorithm s/he should be careful. In similar cases, authors tend to aim for a ~44% acceptance rate per parameter.

[ParadaCarleton]

This is called Gibbs sampling and is implemented in Turing.jl here.