Open mtabone opened 6 years ago
P.S. I'm not sure if this is the proper place for this comment. I'm new to "issues"
More needs to be done here. We are planning two integrate the "admmsource" into the general "source" to make model definitions equivalent regardless of the solver (ADMM or not).
It appears that we re-code the objective function for each source signal in ADMM to be sum square errors with no regularization. Below is a line 217 from CSSS.py, it defines the objective function for each source in CSSS before adding the additional ADMM cost.
Could we replace this with the objective function defined in the class in order to preserve the regularization and cost function? Would this violate something in ADMM?