def setUp(self):
super(HarmoniserWithModelTestCase, self).setUp()
native_measures_1 = self.measures[0:self.number_of_measures / 3]
native_measures_2 = self.measures[
self.number_of_measures / 3:2 * self.number_of_measures / 3]
target_measures = self.measures[2 * self.number_of_measures / 3:]
emsr = EmpiricalMagnitudeScalingRelationship(
native_measures=native_measures_1,
target_measures=target_measures)
self.a_model, _ = emsr.apply_regression_model(
LinearModel)
emsr = EmpiricalMagnitudeScalingRelationship(
native_measures=native_measures_2,
target_measures=target_measures)
self.ya_model, _ = emsr.apply_regression_model(
LinearModel)
def test_regression_with_initial_values(self):
# Assess
A = 0.85
B = 1.03
native_measures = catalogue.MagnitudeMeasure.make_from_lists(
'mb',
np.random.uniform(3., 8.5, 1000),
np.random.uniform(0.02, 0.2, 1000))
target_measures = catalogue.MagnitudeMeasure.make_from_lists(
'Mw', A + B * np.array([m.value for m in native_measures]),
np.random.uniform(0.025, 0.2, 1000))
emsr = EmpiricalMagnitudeScalingRelationship(
native_measures,
target_measures)
# Act
_, output = emsr.apply_regression_model(LinearModel,
initial_values=[0, 1])
# Assert
self.assertTrue(allclose(np.array([A, B]), output.beta))
self.assertTrue(output.res_var < 1e-20)
def test_polynomial_regression(self):
# Assess
A = 0.046
B = 0.556
C = 0.673
native_measures = catalogue.MagnitudeMeasure.make_from_lists(
'mb', np.random.uniform(3., 8.5, 1000),
np.random.uniform(0.02, 0.2, 1000))
target_measures = catalogue.MagnitudeMeasure.make_from_lists(
'Mw',
(C + B * np.array([m.value for m in native_measures]) +
A * (np.array([m.value for m in native_measures]) ** 2.)),
np.random.uniform(0.025, 0.2, 1000))
emsr = EmpiricalMagnitudeScalingRelationship(
native_measures,
target_measures)
# Act
_, output = emsr.apply_regression_model(PolynomialModel,
order=2)
# Assert
self.assertTrue(allclose(np.array([C, B, A]), output.beta))
self.assertTrue(output.res_var < 1e-20)
def perform_regression(self):
"""
Actually, perform the regression analysis.
:returns: a list of n regression outputs where n is the number
of models, and a regression output is a dictionary. In this
dictionary the keys model and output are mapped into a
considered output and its corresponding scipy output result
"""
self._emsr = EmpiricalMagnitudeScalingRelationship.make_from_measures(
self._native_scale, self._target_scale,
self.grouped_measures(), self._selector,
self._mu_strategy)
scipy_outputs = []
for model_class, model_kwargs in self._models:
model, output = self._emsr.apply_regression_model(model_class,
**model_kwargs)
scipy_outputs.append({'model': model,
'output': output})
return scipy_outputs