def test_record_solver_nonlinear_nonlinear_run_once(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NonlinearRunOnce()
self.prob.model.nonlinear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
# No norms so no expected norms
expected_abs_error = 0.0
expected_rel_error = 0.0
expected_solver_residuals = None
expected_solver_output = [
{'name': 'px.x', 'values': [1.]},
{'name': 'pz.z', 'values': [5., 2.]},
{'name': 'd1.y1', 'values': [27.8]},
{'name': 'd2.y2', 'values': [12.27257053]},
{'name': 'obj_cmp.obj', 'values': [30.80000468]},
{'name': 'con_cmp1.con1', 'values': [-24.64]},
{'name': 'con_cmp2.con2', 'values': [-11.72742947]}
]
solver_iteration = json.loads(self.solver_iterations)
self.assertEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertEqual(expected_rel_error, solver_iteration['rel_err'])
self.assertEqual(solver_iteration['solver_residuals'], [])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_record_solver_nonlinear_newton(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
self.prob.model.nonlinear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
expected_solver_output = [
{'name': 'px.x', 'values': [1.0]},
{'name': 'pz.z', 'values': [5., 2.]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'd2.y2', 'values': [12.05848815]},
{'name': 'obj_cmp.obj', 'values': [28.58830817]},
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'con_cmp2.con2', 'values': [-11.94151185]}
]
expected_abs_error = 2.1677810075550974e-10
expected_rel_error = 5.966657077752565e-12
self.assertAlmostEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertAlmostEqual(expected_rel_error, solver_iteration['rel_err'])
self.assertEqual(solver_iteration['solver_residuals'], [])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_only_sysincludes_recorded(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.driver.recording_options['record_desvars'] = False
self.prob.driver.recording_options['record_responses'] = False
self.prob.driver.recording_options['record_objectives'] = False
self.prob.driver.recording_options['record_constraints'] = False
self.prob.driver.recording_options['includes'] = ['*']
self.prob.driver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
driver_iteration_data = json.loads(self.driver_iteration_data)
sysincludes = driver_iteration_data['sysincludes']
self.assertEqual(len(driver_iteration_data['sysincludes']), 2)
self.assertEqual(len(driver_iteration_data['objectives']), 0)
self.assertEqual(len(driver_iteration_data['desvars']), 0)
self.assertEqual(len(driver_iteration_data['constraints']), 0)
self.assertEqual(len(driver_iteration_data['responses']), 0)
def test_record_solver_linear_scipy_iterative_solver(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
# used for analytic derivatives
self.prob.model.nonlinear_solver.linear_solver = ScipyKrylov()
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_abs_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_rel_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_solver_residuals'] = True
self.prob.model.nonlinear_solver.linear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
expected_abs_error = 0.0
expected_rel_error = 0.0
expected_solver_output = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
]
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
self.assertAlmostEqual(0.0, solver_iteration['abs_err'])
self.assertAlmostEqual(0.0, solver_iteration['rel_err'])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_only_desvars_recorded(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.driver.recording_options['record_desvars'] = True
self.prob.driver.recording_options['record_responses'] = False
self.prob.driver.recording_options['record_objectives'] = False
self.prob.driver.recording_options['record_constraints'] = False
self.prob.driver.recording_options['includes'] = []
self.prob.driver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
driver_iteration_data = json.loads(self.driver_iteration_data)
self.driver_iteration_data = None
self.assertTrue({'name': 'px.x', 'values': [1.0]} in driver_iteration_data['desvars'])
self.assertTrue({'name': 'pz.z', 'values': [5.0, 2.0]} in driver_iteration_data['desvars'])
self.assertEqual(driver_iteration_data['responses'], [])
self.assertEqual(driver_iteration_data['objectives'], [])
self.assertEqual(driver_iteration_data['constraints'], [])
def test_record_solver_nonlinear_block_jac(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NonlinearBlockJac()
self.prob.model.nonlinear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
expected_abs_error = 7.234027587097439e-07
expected_rel_error = 1.991112651729199e-08
expected_solver_output = [
{'name': 'px.x', 'values': [1.0]},
{'name': 'pz.z', 'values': [5., 2.]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'd2.y2', 'values': [12.05848815]},
{'name': 'obj_cmp.obj', 'values': [28.58830817]},
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'con_cmp2.con2', 'values': [-11.94151185]}
]
self.assertAlmostEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertAlmostEqual(expected_rel_error, solver_iteration['rel_err'])
self.assertEqual(solver_iteration['solver_residuals'], [])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_record_metadata_values(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.d1.add_recorder(self.recorder)
self.prob.setup(check=False)
run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
self.assertTrue(self.recorded_metadata)
metadata = json.loads(self.metadata)
self.assertEqual(metadata['abs2prom']['input']['d1.z'], 'z')
self.assertEqual(metadata['abs2prom']['input']['d1.x'], 'x')
self.assertEqual(metadata['prom2abs']['input']['z'][0], 'd1.z')
self.assertEqual(metadata['prom2abs']['input']['x'][0], 'd1.x')
def test_record_solver_nonlinear_block_gs(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NonlinearBlockGS()
self.prob.model.nonlinear_solver.add_recorder(self.recorder)
self.prob.model.nonlinear_solver.recording_options['record_solver_residuals'] = True
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
coordinate = [0, 'Driver', (0,), 'root._solve_nonlinear', (0,), 'NonlinearBlockGS', (6, )]
expected_abs_error = 1.31880284470753394998e-10
expected_rel_error = 3.6299074030587596e-12
expected_solver_output = [
{'name': 'px.x', 'values': [1.0]},
{'name': 'pz.z', 'values': [5., 2.]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'd2.y2', 'values': [12.05848815]},
{'name': 'obj_cmp.obj', 'values': [28.58830817]},
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'con_cmp2.con2', 'values': [-11.94151185]}
]
expected_solver_residuals = [
{'name': 'px.x', 'values': [-0]},
{'name': 'pz.z', 'values': [-0., -0.]},
{'name': 'd1.y1', 'values': [1.31880284e-10]},
{'name': 'd2.y2', 'values': [0.]},
{'name': 'obj_cmp.obj', 'values': [0.]},
{'name': 'con_cmp1.con1', 'values': [0.]},
{'name': 'con_cmp2.con2', 'values': [0.]},
]
solver_iteration = json.loads(self.solver_iterations)
self.assertAlmostEqual(solver_iteration['abs_err'], expected_abs_error)
self.assertAlmostEqual(solver_iteration['rel_err'], expected_rel_error)
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
for r in expected_solver_residuals:
self.assert_array_close(r, solver_iteration['solver_residuals'])
def test_record_metadata(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.driver.add_recorder(self.recorder)
self.prob.setup(check=False)
run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
self.assertTrue(self.recorded_metadata)
self.recorded_metadata = False
metadata = json.loads(self.metadata)
self.metadata = None
self.assertEqual(len(metadata['abs2prom']['input']), 11)
self.assertEqual(len(metadata['abs2prom']['output']), 7)
self.assertEqual(len(metadata['prom2abs']['input']), 4)
self.assertEqual(len(metadata['prom2abs']['output']), 7)
def test_implicit_component(self, m):
self.setup_endpoints(m)
from openmdao.core.tests.test_impl_comp import QuadraticLinearize, QuadraticJacVec
group = Group()
group.add_subsystem('comp1', IndepVarComp([('a', 1.0), ('b', 1.0), ('c', 1.0)]))
group.add_subsystem('comp2', QuadraticLinearize())
group.add_subsystem('comp3', QuadraticJacVec())
group.connect('comp1.a', 'comp2.a')
group.connect('comp1.b', 'comp2.b')
group.connect('comp1.c', 'comp2.c')
group.connect('comp1.a', 'comp3.a')
group.connect('comp1.b', 'comp3.b')
group.connect('comp1.c', 'comp3.c')
prob = Problem(model=group)
prob.setup(check=False)
prob['comp1.a'] = 1.
prob['comp1.b'] = -4.
prob['comp1.c'] = 3.
comp2 = prob.model.comp2 # ImplicitComponent
comp2.recording_options['record_metadata'] = False
comp2.add_recorder(self.recorder)
t0, t1 = run_driver(prob)
prob.cleanup()
upload(self.filename, self._accepted_token)
expected_inputs = [
{'name': 'comp2.a', 'values': [1.0]},
{'name': 'comp2.b', 'values': [-4.0]},
{'name': 'comp2.c', 'values': [3.0]}
]
expected_outputs = [{'name': 'comp2.x', 'values': [3.0]}]
expected_residuals = [{'name': 'comp2.x', 'values': [0.0]}]
system_iteration = json.loads(self.system_iterations)
for i in expected_inputs:
self.assert_array_close(i, system_iteration['inputs'])
for r in expected_residuals:
self.assert_array_close(r, system_iteration['residuals'])
for o in expected_outputs:
self.assert_array_close(o, system_iteration['outputs'])
def test_record_solver_linear_direct_solver(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
# used for analytic derivatives
self.prob.model.nonlinear_solver.linear_solver = DirectSolver()
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_abs_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_rel_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_solver_residuals'] = True
self.prob.model.nonlinear_solver.linear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
expected_solver_output = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
{'name': 'd1.y1', 'values': [0.00045069]},
{'name': 'd2.y2', 'values': [-0.00225346]},
{'name': 'obj_cmp.obj', 'values': [0.00045646]},
{'name': 'con_cmp1.con1', 'values': [-0.00045069]},
{'name': 'con_cmp2.con2', 'values': [-0.00225346]},
]
expected_solver_residuals = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [-0., -0.]},
{'name': 'd1.y1', 'values': [0.0]},
{'name': 'd2.y2', 'values': [-0.00229801]},
{'name': 'obj_cmp.obj', 'values': [5.75455956e-06]},
{'name': 'con_cmp1.con1', 'values': [-0.]},
{'name': 'con_cmp2.con2', 'values': [-0.]},
]
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
self.assertAlmostEqual(0.0, solver_iteration['abs_err'])
self.assertAlmostEqual(0.0, solver_iteration['rel_err'])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
for r in expected_solver_residuals:
self.assert_array_close(r, solver_iteration['solver_residuals'])
def test_record_solver(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model._nonlinear_solver.recording_options['record_abs_error'] = True
self.prob.model._nonlinear_solver.recording_options['record_rel_error'] = True
self.prob.model._nonlinear_solver.recording_options['record_solver_residuals'] = True
self.prob.model._nonlinear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
expected_solver_output = [
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'con_cmp2.con2', 'values': [-11.941511849]},
{'name': 'pz.z', 'values': [5.0, 2.0]},
{'name': 'obj_cmp.obj', 'values': [28.588308165]},
{'name': 'd2.y2', 'values': [12.058488150]},
{'name': 'px.x', 'values': [1.0]}
]
expected_solver_residuals = [
{'name': 'con_cmp1.con1', 'values': [0.0]},
{'name': 'd1.y1', 'values': [1.318802844707534e-10]},
{'name': 'con_cmp2.con2', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
{'name': 'obj_cmp.obj', 'values': [0.0]},
{'name': 'd2.y2', 'values': [0.0]},
{'name': 'px.x', 'values': [0.0]}
]
solver_iteration = json.loads(self.solver_iterations)
self.assertAlmostEqual(solver_iteration['abs_err'], 1.31880284470753394998e-10)
self.assertAlmostEqual(solver_iteration['rel_err'], 3.6299074030587596e-12)
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
for r in expected_solver_residuals:
self
def test_distrib_record_driver(self):
size = 100 # how many items in the array
prob = Problem()
prob.model = Group()
prob.model.add_subsystem('des_vars',
IndepVarComp('x', np.ones(size)),
promotes=['x'])
prob.model.add_subsystem('plus',
DistributedAdder(size),
promotes=['x', 'y'])
prob.model.add_subsystem('summer', Summer(size), promotes=['y', 'sum'])
self.recorder.options['record_desvars'] = True
self.recorder.options['record_responses'] = True
self.recorder.options['record_objectives'] = True
self.recorder.options['record_constraints'] = True
prob.driver.add_recorder(self.recorder)
prob.model.add_design_var('x')
prob.model.add_objective('sum')
prob.setup(vector_class=PETScVector, check=False)
prob['x'] = np.ones(size)
t0, t1 = run_driver(prob)
prob.cleanup()
if prob.comm.rank == 0:
coordinate = [0, 'Driver', (0, )]
expected_desvars = {
"des_vars.x": prob['des_vars.x'],
}
expected_objectives = {
"summer.sum": prob['summer.sum'],
}
self.assertDriverIterationDataRecorded(
((coordinate, (t0, t1), expected_desvars, None,
expected_objectives, None, None), ), self.eps)
def test_record_line_search_armijo_goldstein(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
model = self.prob.model
model.nonlinear_solver = NewtonSolver()
model.linear_solver = ScipyKrylov()
model._nonlinear_solver.options['solve_subsystems'] = True
model._nonlinear_solver.options['max_sub_solves'] = 4
ls = model._nonlinear_solver.linesearch = ArmijoGoldsteinLS(bound_enforcement='vector')
# This is pretty bogus, but it ensures that we get a few LS iterations.
ls.options['c'] = 100.0
ls.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
expected_abs_error = 3.49773898733e-9
expected_rel_error = expected_abs_error / 2.9086436370499857e-08
solver_iteration = json.loads(self.solver_iterations)
expected_solver_output = [
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'con_cmp2.con2', 'values': [-11.941511849]},
{'name': 'pz.z', 'values': [5.0, 2.0]},
{'name': 'obj_cmp.obj', 'values': [28.588308165]},
{'name': 'd2.y2', 'values': [12.058488150]},
{'name': 'px.x', 'values': [1.0]}
]
self.assertAlmostEqual(solver_iteration['abs_err'], expected_abs_error)
self.assertAlmostEqual(solver_iteration['rel_err'], expected_rel_error)
self.assertEqual(solver_iteration['solver_residuals'], [])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_only_constraints_recorded(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.driver.recording_options['record_desvars'] = False
self.prob.driver.recording_options['record_responses'] = False
self.prob.driver.recording_options['record_objectives'] = False
self.prob.driver.recording_options['record_constraints'] = True
self.prob.driver.recording_options['includes'] = []
self.prob.driver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
driver_iteration_data = json.loads(self.driver_iteration_data)
if driver_iteration_data['constraints'][0]['name'] == 'con_cmp1.con1':
self.assertAlmostEqual(driver_iteration_data['constraints'][0]['values'][0],
-22.42830237)
self.assertAlmostEqual(driver_iteration_data['constraints'][1]['values'][0],
-11.94151185)
self.assertEqual(driver_iteration_data['constraints'][1]['name'], 'con_cmp2.con2')
self.assertEqual(driver_iteration_data['constraints'][0]['name'], 'con_cmp1.con1')
elif driver_iteration_data['constraints'][0]['name'] == 'con_cmp2.con2':
self.assertAlmostEqual(driver_iteration_data['constraints'][1]['values'][0],
-22.42830237)
self.assertAlmostEqual(driver_iteration_data['constraints'][0]['values'][0],
-11.94151185)
self.assertEqual(driver_iteration_data['constraints'][0]['name'], 'con_cmp2.con2')
self.assertEqual(driver_iteration_data['constraints'][1]['name'], 'con_cmp1.con1')
else:
self.assertTrue(False, 'Driver iteration data did not contain\
the expected names for constraints')
self.assertEqual(driver_iteration_data['desvars'], [])
self.assertEqual(driver_iteration_data['objectives'], [])
self.assertEqual(driver_iteration_data['responses'], [])
def test_record_line_search_bounds_enforce(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
model = self.prob.model
model.nonlinear_solver = NewtonSolver()
model.linear_solver = ScipyKrylov()
model.nonlinear_solver.options['solve_subsystems'] = True
model.nonlinear_solver.options['max_sub_solves'] = 4
ls = model.nonlinear_solver.linesearch = BoundsEnforceLS(bound_enforcement='vector')
ls.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
expected_abs_error = 7.02783609310096e-10
expected_rel_error = 8.078674883382422e-07
expected_solver_output = [
{'name': 'con_cmp1.con1', 'values': [-22.42830237]},
{'name': 'd1.y1', 'values': [25.58830237]},
{'name': 'con_cmp2.con2', 'values': [-11.941511849]},
{'name': 'pz.z', 'values': [5.0, 2.0]},
{'name': 'obj_cmp.obj', 'values': [28.588308165]},
{'name': 'd2.y2', 'values': [12.058488150]},
{'name': 'px.x', 'values': [1.0]}
]
solver_iteration = json.loads(self.solver_iterations)
self.assertAlmostEqual(solver_iteration['abs_err'], expected_abs_error)
self.assertAlmostEqual(solver_iteration['rel_err'], expected_rel_error)
self.assertEqual(solver_iteration['solver_residuals'], [])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_record_solver_linear_linear_run_once(self, m):
self.setup_endpoints(m)
# raise unittest.SkipTest("Linear Solver recording not working yet")
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
# used for analytic derivatives
self.prob.model.nonlinear_solver.linear_solver = LinearRunOnce()
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_abs_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_rel_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_solver_residuals'] = True
self.prob.model.nonlinear_solver.linear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
expected_abs_error = 0.0
expected_rel_error = 0.0
expected_solver_output = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
{'name': 'd1.y1', 'values': [-4.15366975e-05]},
{'name': 'd2.y2', 'values': [-4.10568454e-06]},
{'name': 'obj_cmp.obj', 'values': [-4.15366737e-05]},
{'name': 'con_cmp1.con1', 'values': [4.15366975e-05]},
{'name': 'con_cmp2.con2', 'values': [-4.10568454e-06]},
]
self.assertAlmostEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertAlmostEqual(expected_rel_error, solver_iteration['rel_err'])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_record_solver_linear_block_gs(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
# used for analytic derivatives
self.prob.model.nonlinear_solver.linear_solver = LinearBlockGS()
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_abs_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_rel_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_solver_residuals'] = True
self.prob.model.nonlinear_solver.linear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
expected_abs_error = 9.109083208861876e-11
expected_rel_error = 9.114367543620551e-12
expected_solver_output = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
{'name': 'd1.y1', 'values': [0.00045069]},
{'name': 'd2.y2', 'values': [-0.00225346]},
{'name': 'obj_cmp.obj', 'values': [0.00045646]},
{'name': 'con_cmp1.con1', 'values': [-0.00045069]},
{'name': 'con_cmp2.con2', 'values': [-0.00225346]},
]
self.assertAlmostEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertAlmostEqual(expected_rel_error, solver_iteration['rel_err'])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_only_objectives_recorded(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.driver.recording_options['record_desvars'] = False
self.prob.driver.recording_options['record_responses'] = False
self.prob.driver.recording_options['record_objectives'] = True
self.prob.driver.recording_options['record_constraints'] = False
self.prob.driver.recording_options['includes'] = []
self.prob.driver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
driver_iteration_data = json.loads(self.driver_iteration_data)
self.assertAlmostEqual(driver_iteration_data['objectives'][0]['values'][0], 28.5883082)
self.assertEqual(driver_iteration_data['objectives'][0]['name'], 'obj_cmp.obj')
self.assertEqual(driver_iteration_data['desvars'], [])
self.assertEqual(driver_iteration_data['responses'], [])
self.assertEqual(driver_iteration_data['constraints'], [])
def test_record_solver_linear_block_jac(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.nonlinear_solver = NewtonSolver()
# used for analytic derivatives
self.prob.model.nonlinear_solver.linear_solver = LinearBlockJac()
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_abs_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_rel_error'] = True
self.prob.model.nonlinear_solver.linear_solver.recording_options['record_solver_residuals'] = True
self.prob.model.nonlinear_solver.linear_solver.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
upload(self.filename, self._accepted_token)
solver_iteration = json.loads(self.solver_iterations)
expected_abs_error = 9.947388408259769e-11
expected_rel_error = 4.330301334141486e-08
expected_solver_output = [
{'name': 'px.x', 'values': [0.0]},
{'name': 'pz.z', 'values': [0.0, 0.0]},
{'name': 'd1.y1', 'values': [4.55485639e-09]},
{'name': 'd2.y2', 'values': [-2.27783334e-08]},
{'name': 'obj_cmp.obj', 'values': [-2.28447051e-07]},
{'name': 'con_cmp1.con1', 'values': [2.28461863e-07]},
{'name': 'con_cmp2.con2', 'values': [-2.27742837e-08]},
]
self.assertAlmostEqual(expected_abs_error, solver_iteration['abs_err'])
self.assertAlmostEqual(expected_rel_error, solver_iteration['rel_err'])
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
def test_recording_remote_voi(self):
prob = Problem()
prob.model.add_subsystem('par', ParallelGroup())
prob.model.par.add_subsystem('G1', Mygroup())
prob.model.par.add_subsystem('G2', Mygroup())
prob.model.add_subsystem('Obj', ExecComp('obj=y1+y2'))
prob.model.connect('par.G1.y', 'Obj.y1')
prob.model.connect('par.G2.y', 'Obj.y2')
prob.model.add_objective('Obj.obj')
prob.driver = pyOptSparseDriver()
prob.driver.options['optimizer'] = 'SLSQP'
self.recorder.options['record_desvars'] = True
self.recorder.options['record_responses'] = True
self.recorder.options['record_objectives'] = True
self.recorder.options['record_constraints'] = True
self.recorder.options['system_includes'] = [
'par.G1.Cy.y', 'par.G2.Cy.y'
]
prob.driver.add_recorder(self.recorder)
prob.setup(vector_class=PETScVector)
t0, t1 = run_driver(prob)
prob.cleanup()
# Since the test will compare the last case recorded, just check the
# current values in the problem. This next section is about getting those values
# These involve collective gathers so all ranks need to run this
expected_desvars = prob.driver.get_design_var_values()
expected_objectives = prob.driver.get_objective_values()
expected_constraints = prob.driver.get_constraint_values()
# Determine the expected values for the sysincludes
# this gets all of the outputs but just locally
rrank = prob.comm.rank # root ( aka model ) rank.
rowned = prob.model._owning_rank['output']
# names of sysincl vars on this rank
local_sysinclnames = [
n for n in self.recorder.options['system_includes']
if rrank == rowned[n]
]
# Get values for vars on this rank
inputs, outputs, residuals = prob.model.get_nonlinear_vectors()
# Potential local sysvars are in this
sysvars = outputs._names
# Just get the values for the sysincl vars on this rank
local_sysvars = {c: sysvars[c] for c in local_sysinclnames}
# Gather up the values for all the sysincl vars on all ranks
all_vars = prob.model.comm.gather(local_sysvars, root=0)
if prob.comm.rank == 0:
# Only on rank 0 do we have all the values and only on rank 0
# are we doing the testing.
# The all_vars variable is list of dicts from rank 0,1,... In this case just ranks 0 and 1
dct = all_vars[-1]
for d in all_vars[:-1]:
dct.update(d)
expected_sysincludes = {
'par.G1.Cy.y': dct['par.G1.Cy.y'],
'par.G2.Cy.y': dct['par.G2.Cy.y'],
}
if prob.comm.rank == 0:
coordinate = [0, 'SLSQP', (49, )]
self.assertDriverIterationDataRecorded(
((coordinate,
(t0, t1), expected_desvars, None, expected_objectives,
expected_constraints, expected_sysincludes), ), self.eps)
def test_record_driver_system_solver(self, m):
# Test what happens when all three types are recorded:
# Driver, System, and Solver
self.setup_endpoints(m)
if OPT is None:
raise unittest.SkipTest("pyoptsparse is not installed")
if OPTIMIZER is None:
raise unittest.SkipTest("pyoptsparse is not providing SNOPT or SLSQP")
self.setup_sellar_grouped_model()
self.prob.driver = pyOptSparseDriver()
self.prob.driver.options['optimizer'] = OPTIMIZER
self.prob.driver.opt_settings['ACC'] = 1e-9
# Add recorders
# Driver
self.prob.driver.recording_options['record_metadata'] = True
self.prob.driver.recording_options['record_desvars'] = True
self.prob.driver.recording_options['record_responses'] = True
self.prob.driver.recording_options['record_objectives'] = True
self.prob.driver.recording_options['record_constraints'] = True
self.prob.driver.add_recorder(self.recorder)
# System
pz = self.prob.model.pz # IndepVarComp which is an ExplicitComponent
pz.recording_options['record_metadata'] = True
pz.recording_options['record_inputs'] = True
pz.recording_options['record_outputs'] = True
pz.recording_options['record_residuals'] = True
pz.add_recorder(self.recorder)
# Solver
mda = self.prob.model.mda
mda.nonlinear_solver.recording_options['record_metadata'] = True
mda.nonlinear_solver.recording_options['record_abs_error'] = True
mda.nonlinear_solver.recording_options['record_rel_error'] = True
mda.nonlinear_solver.recording_options['record_solver_residuals'] = True
mda.nonlinear_solver.add_recorder(self.recorder)
self.prob.setup(check=False, mode='rev')
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
# Driver recording test
coordinate = [0, 'SLSQP', (7, )]
expected_desvars = [
{'name': 'pz.z', 'values': self.prob['pz.z']},
{'name': 'px.x', 'values': self.prob['px.x']}
]
expected_objectives = [
{'name': 'obj_cmp.obj', 'values': self.prob['obj_cmp.obj']}
]
expected_constraints = [
{'name': 'con_cmp1.con1', 'values': self.prob['con_cmp1.con1']},
{'name': 'con_cmp2.con2', 'values': self.prob['con_cmp2.con2']}
]
driver_iteration_data = json.loads(self.driver_iteration_data)
for d in expected_desvars:
self.assert_array_close(d, driver_iteration_data['desvars'])
for o in expected_objectives:
self.assert_array_close(o, driver_iteration_data['objectives'])
for c in expected_constraints:
self.assert_array_close(c, driver_iteration_data['constraints'])
# System recording test
expected_inputs = []
expected_outputs = [{'name': 'pz.z', 'values': [1.978467, -1.64641e-13]}]
expected_residuals = [{'name': 'pz.z', 'values': [0.0, 0.0]}]
system_iteration = json.loads(self.system_iterations)
self.assertEqual(expected_inputs, system_iteration['inputs'])
for o in expected_outputs:
self.assert_array_close(o, system_iteration['outputs'])
for r in expected_residuals:
self.assert_array_close(r, system_iteration['residuals'])
# Solver recording test
expected_abs_error = 3.90598e-11
expected_rel_error = 2.0701941158e-06
expected_solver_output = [
{'name': 'mda.d2.y2', 'values': [3.75610598]},
{'name': 'mda.d1.y1', 'values': [3.16]}
]
expected_solver_residuals = [
{'name': 'mda.d2.y2', 'values': [0.0]},
{'name': 'mda.d1.y1', 'values': [0.0]}
]
solver_iteration = json.loads(self.solver_iterations)
np.testing.assert_almost_equal(expected_abs_error, solver_iteration['abs_err'], decimal=5)
np.testing.assert_almost_equal(expected_rel_error, solver_iteration['rel_err'], decimal=5)
for o in expected_solver_output:
self.assert_array_close(o, solver_iteration['solver_output'])
for r in expected_solver_residuals:
self.assert_array_close(r, solver_iteration['solver_residuals'])
def test_record_system(self, m):
self.setup_endpoints(m)
self.setup_sellar_model()
self.prob.model.recording_options['record_inputs'] = True
self.prob.model.recording_options['record_outputs'] = True
self.prob.model.recording_options['record_residuals'] = True
self.prob.model.recording_options['record_metadata'] = True
self.prob.model.add_recorder(self.recorder)
d1 = self.prob.model.d1 # instance of SellarDis1withDerivatives, a Group
d1.add_recorder(self.recorder)
obj_cmp = self.prob.model.obj_cmp # an ExecComp
obj_cmp.add_recorder(self.recorder)
self.prob.setup(check=False)
t0, t1 = run_driver(self.prob)
self.prob.cleanup()
upload(self.filename, self._accepted_token)
system_iterations = json.loads(self.system_iterations)
inputs = [
{'name': 'd1.z', 'values': [5.0, 2.0]},
{'name': 'd1.x', 'values': [1.0]},
{'name': 'd2.z', 'values': [5.0, 2.0]},
{'name': 'd1.y2', 'values': [12.05848815]}
]
outputs = [
{'name': 'd1.y1', 'values': [25.58830237]}
]
residuals = [
{'name': 'd1.y1', 'values': [0.0]}
]
for i in inputs:
self.assert_array_close(i, system_iterations['inputs'])
for o in outputs:
self.assert_array_close(o, system_iterations['outputs'])
for r in residuals:
self.assert_array_close(r, system_iterations['residuals'])
inputs = [
{'name': 'con_cmp2.y2', 'values': [12.058488150624356]},
{'name': 'obj_cmp.y1', 'values': [25.58830237000701]},
{'name': 'obj_cmp.x', 'values': [1.0]},
{'name': 'obj_cmp.z', 'values': [5.0, 2.0]}
]
outputs = [
{'name': 'obj_cmp.obj', 'values': [28.58830816]}
]
residuals = [
{'name': 'obj_cmp.obj', 'values': [0.0]}
]
for i in inputs:
self.assert_array_close(i, system_iterations['inputs'])
for o in outputs:
self.assert_array_close(o, system_iterations['outputs'])
for r in residuals:
self.assert_array_close(r, system_iterations['residuals'])
def test_simple_driver_recording(self, m):
self.setup_endpoints(m)
if OPT is None:
raise unittest.SkipTest("pyoptsparse is not installed")
if OPTIMIZER is None:
raise unittest.SkipTest("pyoptsparse is not providing SNOPT or SLSQP")
prob = Problem()
model = prob.model = Group()
model.add_subsystem('p1', IndepVarComp('x', 50.0), promotes=['*'])
model.add_subsystem('p2', IndepVarComp('y', 50.0), promotes=['*'])
model.add_subsystem('comp', Paraboloid(), promotes=['*'])
model.add_subsystem('con', ExecComp('c = - x + y'), promotes=['*'])
model.suppress_solver_output = True
prob.driver = pyOptSparseDriver()
prob.driver.add_recorder(self.recorder)
prob.driver.recording_options['record_desvars'] = True
prob.driver.recording_options['record_responses'] = True
prob.driver.recording_options['record_objectives'] = True
prob.driver.recording_options['record_constraints'] = True
prob.driver.options['optimizer'] = OPTIMIZER
if OPTIMIZER == 'SLSQP':
prob.driver.opt_settings['ACC'] = 1e-9
model.add_design_var('x', lower=-50.0, upper=50.0)
model.add_design_var('y', lower=-50.0, upper=50.0)
model.add_objective('f_xy')
model.add_constraint('c', upper=-15.0)
prob.setup(check=False)
t0, t1 = run_driver(prob)
prob.cleanup()
upload(self.filename, self._accepted_token)
driver_iteration_data = json.loads(self.driver_iteration_data)
expected_desvars = [
{'name': 'p1.x', 'values': [7.1666666]},
{'name': 'p2.y', 'values': [-7.8333333]}
]
expected_objectives = [
{'name': 'comp.f_xy', 'values': [-27.083333]}
]
expected_constraints = [
{'name': 'con.c', 'values': [-15.0]}
]
for d in expected_desvars:
self.assert_array_close(d, driver_iteration_data['desvars'])
for o in expected_objectives:
self.assert_array_close(o, driver_iteration_data['objectives'])
for c in expected_constraints:
self.assert_array_close(c, driver_iteration_data['constraints'])
