Changes to RBMSymm
Created by: chrisrothUT
This PR integrates RBMSymm into the GCNN infrastructure, as a symmetrized RBM is just a GCNN with one hidden layer and a cosh non-linearity.
This accomplishes a few things 1) RBMSymm is now initialized with variance scaling 2) RBMSymm can enforce parity symmetry 3) RBMSymm works with a character table
As a side note I got rid of the visible bias as it doesn't really do anything besides add a constant. I can add it back if necessary