In this release, I have implemented support for A-optimal in networks where only relative measurements are taken (i.e., only s_{ij} for i!=j are provided). For example, in binding free energy calculations, the users can now construct A-optimal network with only relative binding free energy calculations. To do so, you can simply set the diagonal elements s_{ii} to numpy.inf.
A significant API change is in diffnet.covariance(). It now takes the inverse of pairwise variances as inputs. To update your code, please change
covariance( s, n)
to
covariance( cvxopt.div( n, s**2)).