Fixed recursion issue with mosek fusion. Added input for variable lis…

…t when finding a minimum rank completion.

_scripts/wafr_problems.py

@@ -72,23 +72,37 @@ def process_problems(fname="_examples/mw_loc_3d_small.pkl"):
         pickle.dump(df, file)
 
 
-def test_problem(fname="_examples/mw_loc_3d_small.pkl", id=8):
+def test_problem(
+    fname="_examples/mw_loc_3d_small.pkl",
+    id=8,
+):
     problem, df = load_problem(fname=fname, id=id)
 
     problem.plot_asg(remove_vars=["h"], html="asg.html")
     problem.plot_ctree(html="ctree.html")
 
     # Solve standard SDP
     # X, info, solve_time = solve_sdp_homqcqp(problem, tol=1e-10, verbose=True)
+
     # Solve decomposed SDP
     clq_list, info = solve_dsdp(problem, tol=1e-10, verbose=True)
     Y, ranks, factor_dict = problem.get_mr_completion(clq_list)
 
+    # Check solution
     assert Y.shape[1] == 1, ValueError("Decomposed solution is not rank 1")
+    x = []
+    for varname in df.var_sizes.keys():
+        x.append(factor_dict[varname].flatten())
+    x = np.concatenate(x)
+    np.testing.assert_allclose(
+        x, df.x_gt_init, atol=5e-5, err_msg="solution doesn't match"
+    )
 
 
 if __name__ == "__main__":
-    test_problem(fname="_examples/mw_loc_3d_small.pkl")
+    # test_problem(fname="_examples/mw_loc_3d_small.pkl")
+    # test_problem(fname="_examples/mw_loc_3d.pkl")
     # test_problem(fname="_examples/rangeonlyloc2d_no_const-vel_small.pkl")
+    test_problem(fname="_examples/rangeonlyloc2d_no_const-vel.pkl", id=27)
     # process_problems(fname="_examples/mw_loc_3d_small.pkl")
     # process_problems(fname="_examples/rangeonlyloc2d_no_const-vel_small.pkl")

cert_tools/hom_qcqp.py

@@ -74,6 +74,8 @@ def remove_dependent_constraints(self):
         # Remove dependent constraints
         self.As = [self.As[i] for i in range(len(self.As)) if i not in bad_idx]
 
+        return bad_idx
+
     def get_asg(self, rm_homog=False):
         """Generate Aggregate Sparsity Graph for a given problem. Note that values
         of the matrices are irrelevant.
@@ -88,7 +90,7 @@ def get_asg(self, rm_homog=False):
 
         # Combine cost and constraints
         # NOTE: The numerical values here do not matter, just whether or not an element of the matrix is filled.
-        pmat = self.C
+        pmat = self.C.copy()
         for A in self.As:
             pmat += A
         # build variable dictionaries
@@ -459,13 +461,14 @@ def factor_psd_mat(mat, rank_tol=1e5):
 
         return factor, r
 
-    def get_mr_completion(self, clique_mats, rank_tol=1e5, debug=False):
+    def get_mr_completion(self, clique_mats, var_list=None, rank_tol=1e5, debug=False):
         """Complete a positive semidefinite completable matrix using the
         minimum-rank completion as proposed in:
         Jiang, Xin et al. “Minimum-Rank Positive Semidefinite Matrix Completion with Chordal Patterns and Applications to Semidefinite Relaxations.”
 
         Args:
             clique_mats (list): list of nd-arrays representing the SDP solution (per clique)
+            var_list (list): list of keys corresponding to the desired variable ordering. If None, then the object var_list is used
             rank_tol (_type_, optional): Tolerance for determining rank. Defaults to 1e5.
         """
         r_max = 0  # max rank found
@@ -518,8 +521,10 @@ def get_mr_completion(self, clique_mats, rank_tol=1e5, debug=False):
                     factor_dict[key] = U_Q[inds, :]
 
         # Construct full factor
+        if var_list is None:
+            var_list = self.var_list
         Y = []
-        for varname in self.var_list:
+        for varname in var_list:
             Y.append(factor_dict[varname])
         Y = np.vstack(Y)
         return Y, ranks, factor_dict

cert_tools/sparse_solvers.py

@@ -202,15 +202,13 @@ def get_decomp_fusion_expr(pmat_in):
         # decompose matrix
         mat_decomp = problem.decompose_matrix(pmat_in, decomp_method)
         # add clique components to fusion expression
-        expr = None
+        expr_sum_list = []
         for k, pmat in mat_decomp.items():
             clique = cliques[k]
             mat_k = pmat.get_matrix(variables=clique.var_sizes)
             mat_k_fusion = sparse_to_fusion(mat_k)
-            if expr is None:
-                expr = fu.Expr.dot(mat_k_fusion, cvars[k])
-            else:
-                expr += fu.Expr.dot(mat_k_fusion, cvars[k])
+            expr_sum_list.append(fu.Expr.dot(mat_k_fusion, cvars[k]))
+        expr = fu.Expr.add(expr_sum_list)
         return expr
 
     # OBJECTIVE