def test_no_training_data(self):
krig1 = KrigingSurrogate()
try:
krig1.predict([0.,1.])
except RuntimeError,err:
self.assertEqual(str(err),"KrigingSurrogate has not been trained, so no prediction can be made")
def test_1d_kriging2(self):
x = array([[0.05], [.25], [0.61], [0.95]])
y = array([0.738513784857542,-0.210367746201974,-0.489015457891476,12.3033138316612])
krig1 = KrigingSurrogate()
krig1.train(x,y)
self.assertAlmostEqual(1.18375,krig1.thetas,places=7)
def test_get_uncertain_value(self):
x = array([[0.05], [.25], [0.61], [0.95]])
y = array([0.738513784857542,-0.210367746201974,-0.489015457891476,12.3033138316612])
krig1 = KrigingSurrogate(x,y)
self.assertEqual(krig1.get_uncertain_value(1).mu,NormalDistribution(1.,0.).mu)
self.assertEqual(krig1.get_uncertain_value(1).sigma,NormalDistribution(1.,0.).sigma)
def test_1d_kriging3(self):
"""Test for least squares solver utilization when ill-conditioned"""
x = [[case] for case in linspace(0.,1.,40)]
y = sin(x).flatten()
krig1 = KrigingSurrogate(x,y)
new_x = array([0.5])
pred = krig1.predict(new_x)
self.assertAlmostEqual(8.7709e-09,pred.sigma,places=7)
self.assertAlmostEqual(0.479425538688,pred.mu,places=7)
def test_1d_kriging_predictor(self):
x = array([[0.05], [.25], [0.61], [0.95]])
y = array([0.738513784857542,-0.210367746201974,-0.489015457891476,12.3033138316612])
krig1 = KrigingSurrogate(x,y)
new_x = array([0.5])
pred = krig1.predict(new_x)
self.assertTrue(isinstance(pred,NormalDistribution))
self.assertAlmostEqual(2.5086,pred.sigma,places=3)
self.assertAlmostEqual(-1.37201,pred.mu,places=3)
def test_multi_surrogate_models(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {
'd': KrigingSurrogate(),
'c': LogisticRegression()
}
metamodel.model = Simple()
metamodel.recorder = DumbRecorder()
simple = Simple()
metamodel.a = simple.a = 1.
metamodel.b = simple.b = 2.
metamodel.train_next = True
simple.run()
metamodel.run()
metamodel.a = simple.a = 3.
metamodel.b = simple.b = 4.
metamodel.train_next = True
simple.run()
metamodel.run()
self.assertTrue(isinstance(metamodel.d, NormalDistribution))
self.assertTrue(isinstance(metamodel.c, float))
def test_default_execute(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.default_surrogate = KrigingSurrogate()
metamodel.model = Simple()
metamodel.recorder = DumbRecorder()
simple = Simple()
#import pdb; pdb.set_trace()
metamodel.a = simple.a = 1.
metamodel.b = simple.b = 2.
metamodel.train_next = True
simple.run()
metamodel.run()
metamodel.a = simple.a = 1.
metamodel.b = simple.b = 2.
metamodel.train_next = True
simple.run()
metamodel.run()
self.assertEqual(metamodel.c.getvalue(), 3.)
self.assertEqual(metamodel.d.getvalue(), -1.)
self.assertEqual(metamodel.c.getvalue(), simple.c)
self.assertEqual(metamodel.d.getvalue(), simple.d)
def test_2d_kriging(self):
def bran(x):
y = (x[1]-(5.1/(4.*pi**2.))*x[0]**2.+5.*x[0]/pi-6.)**2.+10.*(1.-1./(8.*pi))*cos(x[0])+10.
return y
x = array([[-2.,0.],[-0.5,1.5],[1.,3.],[8.5,4.5],[-3.5,6.],[4.,7.5],[-5.,9.],[5.5,10.5],
[10.,12.],[7.,13.5],[2.5,15.]])
y = array([bran(case) for case in x])
krig1 = KrigingSurrogate(x,y)
pred = krig1.predict([-2.,0.])
self.assertAlmostEqual(bran(x[0]),pred.mu,places=5)
pred = krig1.predict([5.,5.])
self.assertAlmostEqual(14.513550,pred.sigma,places=2)
self.assertAlmostEqual(18.759264,pred.mu,places=2)
def test_warm_start(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.default_surrogate = KrigingSurrogate()
metamodel.model = Simple()
metamodel.recorder = DumbRecorder()
simple = Simple()
cases = []
metamodel.a = 1.
metamodel.b = 2.
metamodel.train_next = True
metamodel.run()
inputs = [('meta2.a', metamodel.a), ('meta2.b', metamodel.b)]
outputs = [('meta2.c', metamodel.c.mu), ('meta2.d', metamodel.d.mu)]
cases.append(Case(inputs=inputs, outputs=outputs))
metamodel.a = 3.
metamodel.b = 5.
metamodel.train_next = True
metamodel.run()
inputs = [('meta2.a', metamodel.a), ('meta2.b', metamodel.b)]
outputs = [('meta2.c', metamodel.c.mu), ('meta2.d', metamodel.d.mu)]
cases.append(Case(inputs=inputs, outputs=outputs))
case_iter = ListCaseIterator(cases)
metamodel2 = MetaModel()
metamodel2.name = 'meta2'
metamodel2.default_surrogate = KrigingSurrogate()
metamodel2.model = Simple()
metamodel2.recorder = DumbRecorder()
metamodel2.warm_start_data = case_iter
metamodel2.a = simple.a = 1
metamodel2.b = simple.b = 2
metamodel.train_next = True
metamodel2.run()
simple.run()
self.assertEqual(metamodel2.c.getvalue(), 3.)
self.assertEqual(metamodel2.d.getvalue(), -1.)
self.assertEqual(metamodel2.c.getvalue(), simple.c)
self.assertEqual(metamodel2.d.getvalue(), simple.d)
def test_excludes_with_vartrees(self):
self.a.mm.default_surrogate = KrigingSurrogate()
self.a.mm.model = InandOutTree()
self.a.mm.excludes = ['ins', 'outs']
# now try changing the excludes
self.a.mm.excludes = ['outs']
self.assertEqual(self.a.mm.surrogate_input_names(), ['ins.a', 'ins.b'])
self.assertEqual(self.a.mm.surrogate_output_names(), [])
def test_proxy(self):
meta = MetaModel()
meta.default_surrogate = KrigingSurrogate()
meta.model = SimpleProxy()
for i in range(3):
meta.train_next = True
meta.run()
self.assertEqual(len(meta._training_data['f_x']), 3)
def test_multi_surrogate_models_bad_surrogate_dict(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.sur_d = KrigingSurrogate()
try:
metamodel.model = Simple()
except RuntimeError,err:
self.assertEqual("meta: No default surrogate model is defined and the following outputs do not have a surrogate model: ['c']. "
"Either specify default_surrogate, or specify a surrogate model for all "
"outputs.",str(err))
def test_default_surrogate_change(self):
metamodel = MetaModel()
mmins = set(metamodel.list_inputs())
mmouts = set(metamodel.list_outputs())
metamodel.default_surrogate = KrigingSurrogate()
metamodel.model = Simple()
metamodel.default_surrogate = LogisticRegression()
attrs = metamodel.get_attributes(io_only=False)
for s in attrs['Slots']:
self.assertNotEqual(s['name'], 'c')
self.assertNotEqual(s['name'], 'd')
inputs = set(metamodel.list_inputs())
outputs = set(metamodel.list_outputs())
self.assertEquals(inputs - mmins, set(['a', 'b']))
self.assertEquals(outputs - mmouts, set(['c', 'd']))
for i in range(3):
metamodel.train_next = True
metamodel.run()
self.assertTrue(len(metamodel._training_data['c']) == 3)
self.assertTrue(len(metamodel._training_data['d']) == 3)
self.assertTrue(len(metamodel._training_input_history) == 3)
metamodel.includes = ['a', 'b', 'c', 'd']
self.assertTrue(len(metamodel._training_data['c']) == 3)
self.assertTrue(len(metamodel._training_data['d']) == 3)
self.assertTrue(len(metamodel._training_input_history) == 3)
# removing an output should not clobber the rest of the training data
metamodel.includes = ['a', 'b', 'c']
self.assertTrue(len(metamodel._training_data['c']) == 3)
self.assertTrue('d' not in metamodel._training_data)
self.assertTrue(len(metamodel._training_input_history) == 3)
# now put a different model in with the same inputs/outputs
metamodel.model = SimpleMatch()
metamodel.includes = ['a', 'b', 'c', 'd']
inputs = set(metamodel.list_inputs())
outputs = set(metamodel.list_outputs())
self.assertEquals(inputs - mmins, set(['a', 'b']))
self.assertEquals(outputs - mmouts, set(['c', 'd']))
self.assertTrue(len(metamodel._training_data['c']) == 0)
self.assertTrue(len(metamodel._training_data['d']) == 0)
self.assertTrue(len(metamodel._training_input_history) == 0)
# now put a different model in
metamodel.model = Simple2()
metamodel.includes = ['w', 'x', 'y', 'z']
inputs = set(metamodel.list_inputs())
outputs = set(metamodel.list_outputs())
self.assertEquals(inputs - mmins, set(['w', 'x']))
self.assertEquals(outputs - mmouts, set(['y', 'z']))
def test_multi_surrogate_models_bad_surrogate_dict(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {'d': KrigingSurrogate()}
try:
metamodel.model = Simple()
except ValueError, err:
self.assertEqual(
'meta: No default surrogate model was specified. '
'Either specify a default, or specify a surrogate model for all '
'outputs', str(err))
def test_exclude_vartree_leaf(self):
# Should not be allowed in this simplified
# implementation of variable trees in metamodel
self.a.mm.model = InandOutTree()
self.a.mm.default_surrogate = KrigingSurrogate()
try:
self.a.mm.includes = ['ins.a','outs.y']
except Exception as err:
self.assertEqual(str(err),
'mm: Can only include top level variable trees, not leaves')
else:
self.fail('Expected Exception')
def test_excludes(self):
metamodel = MyMetaModel()
metamodel.default_surrogate = KrigingSurrogate()
metamodel.excludes = ['b', 'd']
metamodel.model = Simple()
self.assertEqual(metamodel.surrogate_input_names(), ['a'])
self.assertEqual(metamodel.surrogate_output_names(), ['c'])
# now try changing the excludes
metamodel.excludes = ['a', 'c']
self.assertEqual(metamodel.surrogate_input_names(), ['b'])
self.assertEqual(metamodel.surrogate_output_names(), ['d'])
def test_excludes(self):
metamodel = MyMetaModel()
metamodel.surrogate = {'default': KrigingSurrogate()}
metamodel.excludes = ['a', 'd']
metamodel.model = Simple()
self.assertEqual(metamodel.list_inputs_to_model(), ['b'])
self.assertEqual(metamodel.list_outputs_from_model(), ['c'])
# now try changing the excludes
metamodel.excludes = ['b', 'c']
self.assertEqual(metamodel.list_inputs_to_model(), ['a'])
self.assertEqual(metamodel.list_outputs_from_model(), ['d'])
def test_include_exclude_with_vartrees(self):
self.a.mm.model = InandOutTree()
self.a.mm.default_surrogate = KrigingSurrogate()
self.a.mm.includes = ['ins']
self.a.mm.excludes = ['outs']
try:
self.a.mm.run()
except Exception as err:
self.assertEqual(
str(err), 'mm: includes and excludes are mutually exclusive')
else:
self.fail('Expected Exception')
def _trained_asm(self, avals, bvals):
asm = set_as_top(Assembly())
asm.add('metamodel', MetaModel())
asm.metamodel.surrogate = {'default': KrigingSurrogate()}
asm.metamodel.model = Simple()
asm.metamodel.recorder = DumbRecorder()
asm.driver.workflow.add(['metamodel'])
for a, b in zip(avals, bvals):
asm.metamodel.a = a
asm.metamodel.b = b
asm.metamodel.train_next = 1
asm.metamodel.run()
return asm
def test_comp_error(self):
a = Assembly()
a.add('m', MetaModel())
a.m.surrogate = {'default': KrigingSurrogate()}
a.m.model = DummyError()
a.m.train_next = True
a.driver.workflow.add('m')
try:
a.run()
except RuntimeError as err:
self.assertEqual("m.model: Test Error", str(err))
else:
self.fail("RuntimeError expected")
def _get_assembly(self):
asm = set_as_top(Assembly())
asm.add('metamodel', MetaModel())
asm.add('comp1', Simple())
asm.add('comp2', Simple())
asm.metamodel.surrogate = {'default': KrigingSurrogate()}
asm.metamodel.model = Simple()
asm.metamodel.recorder = DumbRecorder()
asm.driver.workflow.add(['metamodel', 'comp1', 'comp2'])
asm.connect('comp1.c', 'metamodel.a')
asm.connect('comp1.d', 'metamodel.b')
asm.connect('metamodel.c', 'comp2.a')
asm.connect('metamodel.d', 'comp2.b')
return asm
def test_reset_training_data_event(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {'default': KrigingSurrogate()}
metamodel.model = Simple()
metamodel.recorder = DumbRecorder()
simple = Simple()
metamodel.a = 1.
metamodel.b = 2.
metamodel.train_next = True
metamodel.run()
metamodel.a = 2.
metamodel.b = 3.
metamodel.train_next = True
metamodel.run()
def test_model_change(self):
metamodel = MetaModel()
mmins = set(metamodel.list_inputs())
mmouts = set(metamodel.list_outputs())
metamodel.surrogate = {'default': KrigingSurrogate()}
metamodel.model = Simple()
inputs = set(metamodel.list_inputs())
outputs = set(metamodel.list_outputs())
self.assertEquals(inputs - mmins, set(['a', 'b']))
self.assertEquals(outputs - mmouts, set(['c', 'd']))
# now put a different model in
metamodel.model = Simple2()
inputs = set(metamodel.list_inputs())
outputs = set(metamodel.list_outputs())
self.assertEquals(inputs - mmins, set(['w', 'x']))
self.assertEquals(outputs - mmouts, set(['y', 'z']))
def _get_assembly(self, default=True, meta=True):
asm = set_as_top(Assembly())
asm.add('comp1', Simple())
asm.add('comp2', Simple())
if meta:
asm.add('metamodel', MetaModel())
if default:
asm.metamodel.default_surrogate = KrigingSurrogate()
asm.metamodel.model = Simple()
asm.metamodel.recorder = DumbRecorder()
else:
asm.add('metamodel', Simple()) # put a real Simple comp in place of metamodel
asm.driver.workflow.add(['metamodel','comp1','comp2'])
asm.connect('comp1.c','metamodel.a')
asm.connect('comp1.d','metamodel.b')
asm.connect('metamodel.c','comp2.a')
asm.connect('metamodel.d','comp2.b')
return asm
def test_include_exclude(self):
metamodel = MyMetaModel()
metamodel.surrogate = {'default': KrigingSurrogate()}
metamodel.includes = ['a', 'd']
try:
metamodel.excludes = ['b', 'c']
except RuntimeError as err:
self.assertEqual(str(err),
': includes and excludes are mutually exclusive')
else:
self.fail('Expected RuntimeError')
self.assertEqual(metamodel.excludes, [])
metamodel.includes = []
metamodel.excludes = ['b', 'c']
try:
metamodel.includes = ['a', 'd']
except Exception as err:
self.assertEqual(str(err),
': includes and excludes are mutually exclusive')
else:
self.fail('Expected Exception')
self.assertEqual(metamodel.includes, [])
from time import time
import numpy as np
from numpy import pi, cos, sin
from openmdao.lib.surrogatemodels.kriging_surrogate import KrigingSurrogate
np.random.seed(12345)
N = 200
def bran(x):
y = (x[1]-(5.1/(4.*pi**2.))*x[0]**2.+5.*x[0]/pi-6.)**2.+10.*(1.-1./(8.*pi))*cos(x[0])+10.
return y
x = np.random.random((N, 2))*10.0
y = np.array([bran(case) for case in x])
krig1 = KrigingSurrogate()
t0 = time()
krig1.train(x, y)
print 'Training Time elapsed', time() - t0
xx = np.random.random((N, 2))*10.0
t0 = time()
for jj in range(len(xx)):
pred = krig1.predict(xx[jj, :])
print 'predicting Time elapsed', time() - t0
metamodel.name = 'meta'
metamodel.model = Simple()
metamodel.surrogates['d'] = KrigingSurrogate()
try:
metamodel.run()
except RuntimeError, err:
self.assertEqual(
"meta: No default surrogate model is defined and the following outputs do not have a surrogate model: ['c']. "
"Either specify default_surrogate, or specify a surrogate model for all "
"outputs.", str(err))
else:
self.fail('ValueError expected')
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogates['d'] = KrigingSurrogate()
metamodel.includes = ['a', 'b', 'd']
try:
metamodel.model = Simple()
except ValueError, err:
if 'meta: No surrogate was provided for "c". All outputs must have a surrogate' == str(
err):
self.fail(
'should not get a value error for variable c. It is not included in the metamodel'
)
def test_multi_surrogate_models(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.model = Simple()
metamodel.surrogates['d'] = KrigingSurrogate()
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {'d': KrigingSurrogate()}
try:
metamodel.model = Simple()
except ValueError, err:
self.assertEqual(
'meta: No default surrogate model was specified. '
'Either specify a default, or specify a surrogate model for all '
'outputs', str(err))
else:
self.fail('ValueError expected')
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {'d': KrigingSurrogate()}
metamodel.includes = ['a', 'b', 'd']
try:
metamodel.model = Simple()
except ValueError, err:
if 'meta: Dict provided for "surrogates" does not include a value for "c". All outputs must be specified' == str(
err):
self.fail(
'should not get a value error for variable c. It is not included in the metamodel'
)
def test_multi_surrogate_models(self):
metamodel = MetaModel()
metamodel.name = 'meta'
metamodel.surrogate = {
'd': KrigingSurrogate(),
def test_in_and_out_tree_multiple_surrogates(self):
self.a.mm.model = InandOutTree()
self.a.mm.surrogates['outs.x'] = FloatKrigingSurrogate()
self.a.mm.surrogates['outs.y'] = KrigingSurrogate()
self._run_sweep(self.a, 'ins.a', 'ins.b', 'outs.x', 'outs.y')
def test_in_tree_only_multiple_surrogates(self):
self.a.mm.model = InTreeOnly()
self.a.mm.surrogates['x'] = FloatKrigingSurrogate()
self.a.mm.surrogates['y'] = KrigingSurrogate()
self._run_sweep(self.a, 'ins.a', 'ins.b', 'x', 'y')
def configure(self):
self.add('mm', MetaModel())
self.mm.surrogate = {'default': KrigingSurrogate()}
self.mm.model = Dummy()
def configure(self):
self.add('mm', MetaModel())
self.mm.default_surrogate = KrigingSurrogate()
self.mm.model = Dummy()