new_majiq.PsiCoverage.approximate_alpha

property PsiCoverage.approximate_alpha

array(prefix, ec_idx) alpha parameter of approximated bootstrap posterior

In many cases, we operate on the bootstrapped distributions as a single distribution by treating it as a uniform mixture over bootstrap replicates. This mixture is an poorly-behaved model for fixed number of bootstrap replicates as the total coverage increases (the bootstrap replicates behave as atoms). This motivates making a smooth approximation by a single beta distribution. Here, we approximate the beta mixture by matching mean and variance, which we prefer in most cases.