The class mml implements a modified version of the
empirical Bayes method by Gravestock & Held (2017) which was proposed for
borrowing strength from an external study. In their approach, the sharing
weight is found as the maximum of the marginal likelihood of the
weight, given the external data set. This leads, however, to
non-symmetric weights when applied to sharing in basket trials, i.e.
Basket i would not share the information from Basket j as the other way
round. Therefore, a symmetrised version is used, where the mean of the
two weights resulting from sharing in both directions is used.