def test_feature_specification(self):
from openmdao.api import Problem, Group, IndepVarComp, NewtonSolver, ScipyIterativeSolver, ArmijoGoldsteinLS
from openmdao.test_suite.components.implicit_newton_linesearch import ImplCompTwoStates
top = Problem()
top.model = Group()
top.model.add_subsystem('px', IndepVarComp('x', 1.0))
top.model.add_subsystem('comp', ImplCompTwoStates())
top.model.connect('px.x', 'comp.x')
top.model.nonlinear_solver = NewtonSolver()
top.model.nonlinear_solver.options['maxiter'] = 10
top.model.linear_solver = ScipyIterativeSolver()
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS()
ls.options['maxiter'] = 10
top.setup(check=False)
top['px.x'] = 2.0
top['comp.y'] = 0.0
top['comp.z'] = 1.6
top.run_model()
assert_rel_error(self, top['comp.z'], 1.5, 1e-8)
def test_linesearch_vector_bound_enforcement(self):
top = self.top
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(bound_enforcement='vector')
ls.options['c'] = .1
# Setup again because we assigned a new linesearch
top.setup(check=False)
# Test lower bounds: should go to the lower bound and stall
top['px.x'] = 2.0
top['comp.y'] = 0.
top['comp.z'] = 1.6
top.run_model()
for ind in range(3):
assert_rel_error(self, top['comp.z'][ind], [1.5], 1e-8)
# Test upper bounds: should go to the minimum upper bound and stall
top['px.x'] = 0.5
top['comp.y'] = 0.
top['comp.z'] = 2.4
top.run_model()
for ind in range(3):
assert_rel_error(self, top['comp.z'][ind], [2.5], 1e-8)
def test_linesearch_bounds_wall(self):
top = self.top
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='wall')
ls.options['maxiter'] = 10
ls.options['alpha'] = 10.0
# Setup again because we assigned a new linesearch
top.setup(check=False)
# Test lower bound: should go to the lower bound and stall
top['px.x'] = 2.0
top['comp.y'] = 0.0
top['comp.z'] = 1.6
top.run_model()
assert_rel_error(self, top['comp.z'], 1.5, 1e-8)
# Test upper bound: should go to the upper bound and stall
top['px.x'] = 0.5
top['comp.y'] = 0.0
top['comp.z'] = 2.4
top.run_model()
assert_rel_error(self, top['comp.z'], 2.5, 1e-8)
def test_linesearch_wall_bound_enforcement_scalar(self):
top = self.top
top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='scalar')
# Setup again because we assigned a new linesearch
top.setup(check=False)
# Test lower bounds: should stop just short of the lower bound
top['px.x'] = 2.0
top['comp.y'] = 0.
top['comp.z'] = 1.6
top.run_model()
for ind in range(3):
self.assertTrue(1.5 <= top['comp.z'][ind] <= 1.6)
# Test upper bounds: should stop just short of the minimum upper bound
top['px.x'] = 0.5
top['comp.y'] = 0.
top['comp.z'] = 2.4
top.run_model()
for ind in range(3):
self.assertTrue(2.4 <= top['comp.z'][ind] <= self.ub[ind])
def test_linesearch_wall_bound_enforcement_wall(self):
top = self.top
top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='wall')
# Setup again because we assigned a new linesearch
top.setup(check=False)
# Test lower bounds: should go to the lower bound and stall
top['px.x'] = 2.0
top['comp.y'] = 0.
top['comp.z'] = 1.6
top.run_model()
for ind in range(3):
assert_rel_error(self, top['comp.z'][ind], [1.5], 1e-8)
# Test upper bounds: should go to the upper bound and stall
top['px.x'] = 0.5
top['comp.y'] = 0.
top['comp.z'] = 2.4
top.run_model()
for ind in range(3):
assert_rel_error(self, top['comp.z'][ind], [self.ub[ind]], 1e-8)
def test_linesearch_bounds_scalar(self):
top = self.top
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='scalar')
ls.options['maxiter'] = 10
ls.options['alpha'] = 10.0
# Setup again because we assigned a new linesearch
top.setup(check=False)
# Test lower bound: should stop just short of the lower bound
top['px.x'] = 2.0
top['comp.y'] = 0.0
top['comp.z'] = 1.6
top.run_model()
self.assertTrue(1.5 <= top['comp.z'] <= 1.6)
# Test lower bound: should stop just short of the upper bound
top['px.x'] = 0.5
top['comp.y'] = 0.0
top['comp.z'] = 2.4
top.run_model()
self.assertTrue(2.4 <= top['comp.z'] <= 2.5)
def test_deep_analysis_error_iprint(self):
class ImplCompTwoStatesAE(ImplicitComponent):
def setup(self):
self.add_input('x', 0.5)
self.add_output('y', 0.0)
self.add_output('z', 2.0, lower=1.5, upper=2.5)
self.maxiter = 10
self.atol = 1.0e-12
self.declare_partials(of='*', wrt='*')
self.counter = 0
def apply_nonlinear(self, inputs, outputs, residuals):
"""
Don't solve; just calculate the residual.
"""
x = inputs['x']
y = outputs['y']
z = outputs['z']
residuals['y'] = y - x - 2.0 * z
residuals['z'] = x * z + z - 4.0
self.counter += 1
if self.counter > 5 and self.counter < 11:
raise AnalysisError('catch me')
def linearize(self, inputs, outputs, jac):
"""
Analytical derivatives.
"""
# Output equation
jac[('y', 'x')] = -1.0
jac[('y', 'y')] = 1.0
jac[('y', 'z')] = -2.0
# State equation
jac[('z', 'z')] = -inputs['x'] + 1.0
jac[('z', 'x')] = -outputs['z']
top = Problem()
top.model = Group()
top.model.add_subsystem('px', IndepVarComp('x', 7.0))
sub = top.model.add_subsystem('sub', Group())
sub.add_subsystem('comp', ImplCompTwoStatesAE())
top.model.connect('px.x', 'sub.comp.x')
top.model.nonlinear_solver = NewtonSolver()
top.model.nonlinear_solver.options['maxiter'] = 2
top.model.nonlinear_solver.options['solve_subsystems'] = True
top.model.linear_solver = ScipyKrylov()
sub.nonlinear_solver = NewtonSolver()
sub.nonlinear_solver.options['maxiter'] = 2
sub.linear_solver = ScipyKrylov()
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='wall')
ls.options['maxiter'] = 5
ls.options['alpha'] = 10.0
ls.options['retry_on_analysis_error'] = True
ls.options['c'] = 10000.0
top.setup(check=False)
top.set_solver_print(level=2)
stdout = sys.stdout
strout = StringIO()
sys.stdout = strout
try:
top.run_model()
finally:
sys.stdout = stdout
output = strout.getvalue().split('\n')
self.assertTrue(output[26].startswith('| LS: AG 5'))
def test_analysis_error(self):
class ParaboloidAE(ExplicitComponent):
""" Evaluates the equation f(x,y) = (x-3)^2 + xy + (y+4)^2 - 3
This version raises an analysis error if x < 2.0
The AE in ParaboloidAE stands for AnalysisError."""
def __init__(self):
super(ParaboloidAE, self).__init__()
self.fail_hard = False
def setup(self):
self.add_input('x', val=0.0)
self.add_input('y', val=0.0)
self.add_output('f_xy', val=0.0)
self.declare_partials(of='*', wrt='*')
def compute(self, inputs, outputs):
"""f(x,y) = (x-3)^2 + xy + (y+4)^2 - 3
Optimal solution (minimum): x = 6.6667; y = -7.3333
"""
x = inputs['x']
y = inputs['y']
if x < 1.75:
raise AnalysisError('Try Again.')
outputs['f_xy'] = (x - 3.0)**2 + x * y + (y + 4.0)**2 - 3.0
def compute_partials(self, inputs, partials):
""" Jacobian for our paraboloid."""
x = inputs['x']
y = inputs['y']
partials['f_xy', 'x'] = 2.0 * x - 6.0 + y
partials['f_xy', 'y'] = 2.0 * y + 8.0 + x
top = Problem()
top.model = Group()
top.model.add_subsystem('px', IndepVarComp('x', 1.0))
top.model.add_subsystem('comp', ImplCompTwoStates())
top.model.add_subsystem('par', ParaboloidAE())
top.model.connect('px.x', 'comp.x')
top.model.connect('comp.z', 'par.x')
top.model.nonlinear_solver = NewtonSolver()
top.model.nonlinear_solver.options['maxiter'] = 1
top.model.linear_solver = ScipyKrylov()
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='vector')
ls.options['maxiter'] = 10
ls.options['alpha'] = 1.0
top.set_solver_print(level=0)
top.setup(check=False)
# Test lower bound: should go as far as it can without going past 1.75 and triggering an
# AnalysisError. It doesn't do a great job, so ends up at 1.8 instead of 1.75
top['px.x'] = 2.0
top['comp.y'] = 0.0
top['comp.z'] = 2.1
top.run_model()
assert_rel_error(self, top['comp.z'], 1.8, 1e-8)
# Test the behavior with the switch turned off.
top = Problem()
top.model = Group()
top.model.add_subsystem('px', IndepVarComp('x', 1.0))
top.model.add_subsystem('comp', ImplCompTwoStates())
top.model.add_subsystem('par', ParaboloidAE())
top.model.connect('px.x', 'comp.x')
top.model.connect('comp.z', 'par.x')
top.model.nonlinear_solver = NewtonSolver()
top.model.nonlinear_solver.options['maxiter'] = 1
top.model.linear_solver = ScipyKrylov()
ls = top.model.nonlinear_solver.linesearch = ArmijoGoldsteinLS(
bound_enforcement='vector')
ls.options['maxiter'] = 10
ls.options['alpha'] = 1.0
ls.options['retry_on_analysis_error'] = False
top.set_solver_print(level=0)
top.setup(check=False)
top['px.x'] = 2.0
top['comp.y'] = 0.0
top['comp.z'] = 2.1
with self.assertRaises(AnalysisError) as context:
top.run_model()
self.assertEqual(str(context.exception), 'Try Again.')