fixing issue with sparsity in projection

cvxpy/expressions/leaf.py

@@ -422,7 +422,9 @@ def project(self, val):
                 w[bad] = 0
             return (V * w).dot(V.T)
         elif self.attributes['sparsity']:
-            return np.where(val[self.sparse_idx], val, 0)
+            new_val = np.zeros(self.shape)
+            new_val[self.sparse_idx] = val
+            return new_val
         else:
             return val
 

cvxpy/reductions/cvx_attr2constr.py

@@ -155,9 +155,9 @@ def apply(self, problem):
                     id2new_var[var.id] = sparse_var
                     row_idx = np.ravel_multi_index(var.sparse_idx, var.shape)
                     col_idx = np.arange(n)
-                    matrix = Constant(sp.csc_matrix((np.ones(n), (row_idx, col_idx)),
+                    coeff_matrix = Constant(sp.csc_matrix((np.ones(n), (row_idx, col_idx)),
                                                     shape=(np.prod(var.shape, dtype=int), n)))
-                    obj = reshape(matrix @ sparse_var, var.shape)
+                    obj = reshape(coeff_matrix @ sparse_var, var.shape)
                 elif var.attributes['diag']:
                     diag_var = Variable(var.shape[0], var_id=var.id, **new_attr)
                     diag_var.set_variable_of_provenance(var)

cvxpy/tests/test_domain.py

@@ -194,7 +194,21 @@ def test_matrix_frac(self) -> None:
 class TestAttributes():
     def test_sparsity_pattern(self):
         X = cp.Variable((3, 3), sparsity=[(0, 0), (1, 1), (2, 2)])
+        target = np.eye(3)
         assert X.sparse_idx == ((0, 1, 2), (0, 1, 2))
+        prob = cp.Problem(cp.Minimize(0), [X == target])
+        assert prob.solve(verbose=True) == 0
+        assert X.value is not None
+
+    def test_diag_sparsity(self):
+        X = cp.Variable((3, 3), diag=True)
         prob = cp.Problem(cp.Minimize(cp.norm(X, 'fro')))
         assert prob.solve(verbose=True) == 0
         assert X.value is not None
+
+    def test_sparsity_pattern_similar_diag(self):
+        X = cp.Variable((3, 3), sparsity=[(0, 0), (1, 1), (2, 2)])
+        assert X.sparse_idx == ((0, 1, 2), (0, 1, 2))
+        prob = cp.Problem(cp.Minimize(cp.norm(X, 'fro')))
+        assert prob.solve(verbose=True) == 0
+        assert X.value is not None