We present an implementation of the recently proposed imperfect pairing (IP) and generalized valence bond restricted coupled cluster (GVB-RCC) methods. Our algorithm centers on repeated construction of Coulomb and exchange matrices. These operations are the computational bottleneck, scaling with the third power of system size for large systems. Robust optimization of the valence orbitals is attained using a geometrically consistent form of direct minimization. Analytic gradients of the IP and GVB-RCC energies are also obtained by a simple modification of the energy optimization scheme. As an illustration of the potential of these new methods, we use IP to compute the equilibrium geometry and energetics of a Si9H12 cluster that is a crude model for silicon dimerization on the Si(001) surface. We thus demonstrate a valuable role for IP and GVB-RCC as a diagnostic for the accuracy of reduced active space calculations as compared to their full valence analogs.