Merge pull request #17 from cvxgrp/dev

Dev

gfosd/components/basis_constraints.py

@@ -19,7 +19,13 @@
 class Basis(GraphComponent):
     def __init__(self, basis, penalty=None, *args, **kwargs):
         self._basis = basis
+        # penalty can be None, an atom name (e.g. 'sum_square' or 'abs'), or PSD matrix (2D numpy array)
         self._penalty = penalty
+        if isinstance(penalty, np.ndarray) or isinstance(penalty, sp.spmatrix):
+            self._penalty = 'matrix'
+            self._pmat = penalty
+        else:
+            self._ndim = None
         super().__init__(*args, **kwargs)
         self._has_helpers = True
 
@@ -40,19 +46,23 @@ def _make_B(self):
         self._B = self._basis * -1
 
     def _make_g(self, size):
-        if (self._penalty is None) or (self._penalty == 'sum_square'):
+        if (self._penalty is None) or (self._penalty == 'sum_square') or (self._penalty == 'matrix'):
             g = []
         else:
+            # typically 'abs', 'huber', or 'quantile'
             g = [{'g': self._penalty,
                          'args': {'weight': self.weight},
                          'range': (0, size)}]
         return g
 
     def _make_P(self, size):
-        if (self._penalty is None) or (self._penalty != 'sum_square'):
-            P = sp.dok_matrix(2 * (size,))
-        else:
+        if self._penalty == 'matrix':
+            P = sp.dia_matrix(self._pmat)
+            P = P.power(2)
+        elif np.all(self._penalty == 'sum_square'):
             P = self.weight * sp.eye(size)
+        else:
+            P = sp.dok_matrix(2 * (size,))
         return P
 
 class Periodic(Basis):

gfosd/problem.py

@@ -196,7 +196,15 @@ def _solve_cvx(self, data, solver, **solver_kwargs):
                 return
         objective = cvx.Minimize(cost)
         cvx_prob = cvx.Problem(objective, constraints)
-
+        if solver == "CLARABEL":
+            if "eps_rel" in solver_kwargs.keys():
+                er = solver_kwargs["eps_rel"]
+                del solver_kwargs["eps_rel"]
+                solver_kwargs["tol_gap_rel"] = er
+            if "eps_abs" in solver_kwargs.keys():
+                ea = solver_kwargs["eps_abs"]
+                del solver_kwargs["eps_abs"]
+                solver_kwargs["tol_gap_abs"] = ea
         cvx_prob.solve(solver=solver, **solver_kwargs)
         self._cvx_obj = cvx_prob
         self.objective_value = cvx_prob.value

requirements.txt

@@ -4,3 +4,4 @@ cvxpy
 matplotlib
 scikit-learn
 qss
+clarabel