def apply(self, problem: ParamConeProg) -> Tuple[Dict, Dict]:
"""Returns a new problem and data for inverting the new solution."""
from ortools.linear_solver import linear_solver_pb2
# Create data and inv_data objects
data = {}
inv_data = {self.VAR_ID: problem.x.id}
if not problem.formatted:
problem = self.format_constraints(problem, None)
data[s.PARAM_PROB] = problem
data[self.DIMS] = problem.cone_dims
constr_map = problem.constr_map
inv_data["constraints"] = constr_map[Zero] + constr_map[NonNeg]
# Min c'x + d such that Ax + b = s, s \in cones.
c, d, A, b = problem.apply_parameters()
A = csr_matrix(A)
data["num_constraints"], data["num_vars"] = A.shape
# TODO: Switch to a vectorized model-building interface when one is
# available in OR-Tools.
model = linear_solver_pb2.MPModelProto()
model.objective_offset = d.item() if isinstance(d, np.ndarray) else d
for var_index, obj_coef in enumerate(c):
var = linear_solver_pb2.MPVariableProto(
objective_coefficient=obj_coef, name="x_%d" % var_index)
model.variable.append(var)
for row_index in range(A.shape[0]):
constraint = linear_solver_pb2.MPConstraintProto(
name="constraint_%d" % row_index)
start = A.indptr[row_index]
end = A.indptr[row_index + 1]
for nz_index in range(start, end):
col_index = A.indices[nz_index]
coeff = A.data[nz_index]
constraint.var_index.append(col_index)
constraint.coefficient.append(coeff)
if row_index < problem.cone_dims.zero:
# a'x + b == 0
constraint.lower_bound = -b[row_index]
constraint.upper_bound = -b[row_index]
else:
# a'x + b >= 0
constraint.lower_bound = -b[row_index]
model.constraint.append(constraint)
data[self.MODEL_PROTO] = model
return data, inv_data
def test_proto(self):
input_proto = linear_solver_pb2.MPModelProto()
text_format.Merge(TEXT_MODEL, input_proto)
solver = pywraplp.Solver('solveFromProto',
pywraplp.Solver.CBC_MIXED_INTEGER_PROGRAMMING)
# For now, create the model from the proto by parsing the proto
errors = solver.LoadModelFromProto(input_proto)
self.assertFalse(errors)
solver.Solve()
# Fill solution
solution = linear_solver_pb2.MPSolutionResponse()
solver.FillSolutionResponseProto(solution)
self.assertEqual(solution.objective_value, 3.0)
self.assertEqual(solution.variable_value[0], 1.0)
self.assertEqual(solution.variable_value[1], 1.0)
self.assertEqual(solution.best_objective_bound, 3.0)