Scale problem in ADMM / CG about scico

[lukepfister]

Does this only happen with LinearSubproblemSolver? Does it happen with the nonlinear or circulant solvers, too? If it's just LinearSubproblemSolver, I think the problem is that the scaling is being used twice. First as admm.f.scale here

scico/scico/admm.py

Line 188 in 4b5cec4

lhs_op = lhs_op + 2.0 * admm.f.scale * admm.f.hessian

but that scaling is also accounted for within admm.f.hessian:

scico/scico/loss.py

Lines 161 to 162 in 4b5cec4

	if isinstance(self.A, linop.LinearOperator):
	return 2 * self.scale * self.A.gram_op

[tbalke]

@lukepfister Yes, that did it!! Will create a PR.

[bwohlberg]

Resolved by #45