def test_simple(self):
prob = om.Problem()
model = prob.model
interp = om.MetaModelSemiStructuredComp(method='lagrange2',
training_data_gradients=True)
interp.add_input('x', data_x)
interp.add_input('y', data_y)
interp.add_output('f', data_values)
# Sneak in a multi-output case.
interp.add_output('g', 2.0 * data_values)
model.add_subsystem('interp', interp)
prob.setup(force_alloc_complex=True)
prob.set_val('interp.x', np.array([3.1]))
prob.set_val('interp.y', np.array([2.75]))
prob.run_model()
assert_near_equal(prob.get_val('interp.f'), 3.39415716, 1e-7)
assert_near_equal(prob.get_val('interp.g'), 2.0 * 3.39415716, 1e-7)
def test_list_input(self):
x = [1.0, 1.0, 2.0, 2.0, 2.0]
y = [1.0, 2.0, 1.0, 2.0, 3.0]
f = [1.0, 2.5, 1.5, 4.0, 4.5]
comp = om.MetaModelSemiStructuredComp(method='slinear',
training_data_gradients=True,
extrapolate=False)
comp.add_input('x', x)
comp.add_input('y', y)
comp.add_output('f', f)
prob = om.Problem()
model = prob.model
model.add_subsystem('comp', comp)
prob.setup()
prob.set_val('comp.x', 1.5)
prob.set_val('comp.y', 1.5)
prob.run_model()
f = prob.get_val('comp.f')
assert_near_equal(f, 2.25)
# Attempt internal extrapolation.
prob.set_val('comp.x', 1.5)
prob.set_val('comp.y', 2.5)
msg = "'comp' <class MetaModelSemiStructuredComp>: Error interpolating output 'f' because input 'comp.y' required extrapolation while interpolating dimension 2, where its value '2.5' exceeded the range ('[1.]', '[2.]')"
with self.assertRaises(om.AnalysisError) as cm:
prob.run_model()
self.assertEqual(str(cm.exception), msg)
def test_vectorized_akima(self):
# Test using the model we used for the Structured metamodel.
prob = om.Problem()
model = prob.model
ivc = om.IndepVarComp()
mapdata = SampleMap()
params = mapdata.param_data
x, y, _ = params
outs = mapdata.output_data
z = outs[0]
ivc.add_output('x',
np.array([x['default'], x['default'], x['default']]),
units=x['units'])
ivc.add_output('y',
np.array([y['default'], y['default'], y['default']]),
units=x['units'])
ivc.add_output('z',
np.array([z['default'], z['default'], z['default']]),
units=x['units'])
model.add_subsystem('des_vars', ivc, promotes=["*"])
comp = om.MetaModelSemiStructuredComp(method='akima',
extrapolate=True,
training_data_gradients=True,
vec_size=3)
comp._no_check_partials = False # override skipping of check_partials
# Convert to the flat table format.
grid = np.array(
list(
itertools.product(*[
params[0]['values'], params[1]['values'], params[2]
['values']
])))
j = 0
for param in params:
comp.add_input(param['name'], grid[:, j])
j += 1
for out in outs:
comp.add_output(out['name'], outs[0]['values'].flatten())
model.add_subsystem('comp', comp, promotes=["*"])
prob.setup(force_alloc_complex=True)
prob['x'] = np.array([1.0, 10.0, 90.0])
prob['y'] = np.array([0.75, 0.81, 1.2])
prob['z'] = np.array([-1.7, 1.1, 2.1])
prob.run_model()
partials = prob.check_partials(method='cs', out_stream=None)
assert_check_partials(partials, rtol=1e-10)
def test_basic(self):
# Tests that semi structured grids load without error.
grid = np.array([
[1.0, 5.0, 8.0],
[1.0, 5.0, 9.0],
[1.0, 5.0, 10.0],
[1.0, 5.0, 20.0],
[1.0, 5.3, 8.0],
[1.0, 5.3, 9.0],
[1.0, 5.3, 10.0],
[1.0, 5.3, 20.0],
[1.0, 5.6, 8.0],
[1.0, 5.6, 9.0],
[1.0, 5.6, 10.0],
[1.0, 5.6, 20.0],
[1.0, 6.0, 8.0],
[1.0, 6.0, 9.0],
[1.0, 6.0, 10.0],
[1.0, 6.0, 20.0],
[2.0, 7.0, 13.0],
[2.0, 7.0, 14.0],
[2.0, 7.0, 15.0],
[2.0, 7.0, 16.0],
[2.0, 8.0, 13.0],
[2.0, 8.0, 14.0],
[2.0, 8.0, 15.0],
[2.0, 8.0, 16.0],
[2.0, 8.5, 13.0],
[2.0, 8.5, 14.0],
[2.0, 8.5, 15.0],
[2.0, 8.5, 16.0],
[2.0, 9.0, 13.0],
[2.0, 9.0, 14.0],
[2.0, 9.0, 15.0],
[2.0, 9.0, 16.0],
])
values = 15.0 + 2 * np.random.random(32)
prob = om.Problem()
model = prob.model
interp = om.MetaModelSemiStructuredComp(vec_size=3,
training_data_gradients=True)
interp.add_input('x', training_data=grid[:, 0])
interp.add_input('y', training_data=grid[:, 1])
interp.add_input('z', training_data=grid[:, 2])
interp.add_output('f', training_data=values)
model.add_subsystem('interp', interp)
prob.setup(force_alloc_complex=True)
prob.run_model()
viz = MetaModelVisualization(interp)
def test_lagrange3_edge_extrapolation_detection_bug(self):
import itertools
import numpy as np
import openmdao.api as om
grid = np.array(
[[0, 0],
[0, 1],
[0, 2],
[0, 3],
[0, 4],
[1, 0],
[1, 1],
[1, 2],
[1, 3],
[1, 4],
[2, 0],
[2, 1],
[2, 2],
[2, 3],
[2, 4],
[2, 5],
[3, 0],
[3, 1],
[3, 2],
[3, 3],
[3, 4],
[3, 5],
[4, 0],
[4, 1],
[4, 2],
[4, 3],
[4, 4],
[4, 5]])
values = 15.0 + 2 * np.random.random(grid.shape[0])
prob = om.Problem()
model = prob.model
interp = om.MetaModelSemiStructuredComp(method='lagrange3')
interp.add_input('x', training_data=grid[:, 0])
interp.add_input('y', training_data=grid[:, 1])
interp.add_output('f', training_data=values)
model.add_subsystem('interp', interp)
prob.setup()
prob.set_val('interp.x', 2.5)
prob.set_val('interp.y', 4.5)
# Should run without an Indexerror.
prob.run_model()
def test_dynamic_training(self):
p1 = np.linspace(0, 100, 25)
p2 = np.linspace(-10, 10, 5)
p3 = np.linspace(0, 1, 10)
P1, P2, P3 = np.meshgrid(p1, p2, p3, indexing='ij')
P1 = P1.ravel()
P2 = P2.ravel()
P3 = P3.ravel()
class TableGen(om.ExplicitComponent):
def setup(self):
self.add_input('k', 1.0)
self.add_output('values', np.zeros(len(P1)))
self.declare_partials('values', 'k')
def compute(self, inputs, outputs):
k = inputs['k']
outputs['values'] = np.sqrt(P1) + P2 * P3 * k
def compute_partials(self, inputs, partials):
partials['values', 'k'] = P2 * P3
prob = om.Problem()
model = prob.model
model.add_subsystem('tab', TableGen())
interp = om.MetaModelSemiStructuredComp(method='lagrange3', training_data_gradients=True)
interp.add_input('p1', P1)
interp.add_input('p2', P2)
interp.add_input('p3', P3)
interp.add_output('f')
model.add_subsystem('comp', interp)
model.connect('tab.values', 'comp.f_train')
prob.setup(force_alloc_complex=True)
prob.set_val('comp.p1', 55.12)
prob.set_val('comp.p2', -2.14)
prob.set_val('comp.p3', 0.323)
prob.run_model()
# we can verify all gradients by checking against finite-difference
totals = prob.check_totals(of='comp.f', wrt=['tab.k', 'comp.p1', 'comp.p2', 'comp.p3'],
method='cs', out_stream=None);
assert_near_equal(totals['comp.f', 'tab.k']['abs error'][0], 0.0, tolerance=1e-10)
def test_error_no_training_data(self):
x = np.array([1.0, 1.0, 2.0, 2.0])
y = np.array([1.0, 2.0, 1.0, 2.0])
comp = om.MetaModelSemiStructuredComp(method='akima')
comp.add_input('x', x)
comp.add_input('y', y)
msg = "Training data is required for output 'f'."
with self.assertRaisesRegex(ValueError, msg):
comp.add_output('f')
def test_detect_local_extrapolation(self):
# Tests that we detect when any of our points we are using for interpolation are being extrapolated from
# somewhere else in the semi-structured grid, so that we can adjust our points (if we can.)
# This test is set up so that if we aren't actively doing this, lagrange2 and lagrange3 will compute
# large values near the ends. Akima seems to already be robust to this, and didn't require any changes.
# 8x8 block
u = np.arange(24)
v = np.arange(8)
grid = np.empty((192, 2))
grid[:, 0] = np.repeat(u, 8)
grid[:64, 1] = np.tile(v, 8) + 8
grid[64:128, 1] = np.tile(v, 8)
grid[128:, 1] = np.tile(v, 8) + 8
values = np.empty((192, ))
values[:64] = 1e8 * (6.0 + 5.0 * np.sin(.02 * grid[:64, 0]) +
np.sin(.03 * grid[:64, 1]))
values[64:128] = (6.0 + 5.0 * np.sin(.02 * grid[64:128, 0]) +
np.sin(.03 * grid[64:128, 1]))
values[128:] = 1e8 * (6.0 + 5.0 * np.sin(.02 * grid[128:, 0]) +
np.sin(.03 * grid[128:, 1]))
expected = np.array([
6.91181637, 7.01019418, 7.1081943, 7.20577754, 7.30290486,
7.39953742, 7.49563655
])
for method in ['slinear', 'lagrange2', 'lagrange3', 'akima']:
prob = om.Problem()
model = prob.model
interp = om.MetaModelSemiStructuredComp(method=method, vec_size=7)
interp.add_input('x', grid[:, 0])
interp.add_input('y', grid[:, 1])
interp.add_output('f', values)
model.add_subsystem('interp', interp)
prob.setup()
prob.set_val('interp.x',
np.array([8.5, 9.5, 10.5, 11.5, 12.5, 13.5, 14.5]))
prob.set_val('interp.y',
np.array([2.2, 2.2, 2.2, 2.2, 2.2, 2.2, 2.2]))
prob.run_model()
assert_near_equal(prob.get_val('interp.f'), expected, 1e-3)
def test_error_dim(self):
x = np.array([1.0, 1.0, 2.0, 2.0])
y = np.array([1.0, 2.0, 1.0, 2.0])
f = np.array([1.0, 2.0, 3.0])
comp = om.MetaModelSemiStructuredComp(method='akima')
comp.add_input('x', x)
comp.add_input('y', y)
comp.add_output('f', f)
prob = om.Problem()
model = prob.model
model.add_subsystem('comp', comp)
msg = "Size mismatch: training data for 'f' is length 3, but" + \
f" data for 'x' is length 4."
with self.assertRaisesRegex(ValueError, msg):
prob.setup()