Posterior over ratings from accumulated log-likelihoods

A uniform prior over the available networks, so the posterior is the softmax of the log-likelihoods. Computed by subtracting the maximum first, because the log-likelihood of a long game is a large negative number and exp() of it underflows to zero.

Usage

rating_posterior(loglik)

Arguments

loglik

A named numeric vector of total log-likelihoods per rating.

Value

A named numeric vector summing to 1, or all NA if nothing is known.