v1.2

Updates:

• Now you can use --enforece-nonnegative-connectivity-matrix option to enforce that the connectivity matrix remains nonnegative, meaning that all the "spring constants" are either zero or positive. Turning on this option typically result in worse "fitting" to the data. However, it has an advantage if you want to do perturbation such as deletion, inversion, etc on the result connectivity matrix because it won't lead to a non-valid connectivity matrix (being not negative semi-definite)
• Now you can use -m DI to enable the direct inversion method for "fitting" to the data. This will only work if the target distance matrix is a Euclidean matrix or close to a Euclidean matrix. If the distance matrix is computed directly from imaging's (x,y,z) data, this method will naturally work. If the input is the contact map instead, typically this method won't work because the inferred target distance map is often very non-Euclidean. When it does work, it still has one drawback which is that the connectivity matrix obtained is not regularized. Its L2 norm can be very large. Depending on the context, this may or may not be an issue.