def test_plot_lsa(self):
figure_name = "LSAPlot"
hypo_builder = HypothesisBuilder(config=self.config).set_name(figure_name)
lsa_hypothesis = hypo_builder.build_lsa_hypothesis()
mc = ModelConfigurationBuilder().build_model_from_E_hypothesis(lsa_hypothesis, numpy.array([1]))
figure_file = os.path.join(self.config.out.FOLDER_FIGURES, figure_name + ".png")
assert not os.path.exists(figure_file)
self.plotter.plot_lsa(lsa_hypothesis, mc, True, None, region_labels=numpy.array(["a"]), title="")
assert not os.path.exists(figure_file)
def run_lsa(self, disease_hypothesis, model_configuration):
jacobian = self._compute_jacobian(model_configuration)
# Perform eigenvalue decomposition
eigen_values, eigen_vectors = numpy.linalg.eig(jacobian)
sorted_indices = numpy.argsort(eigen_values, kind='mergesort')
self.eigen_values = eigen_values[sorted_indices]
self.eigen_vectors = eigen_vectors[:, sorted_indices]
self._ensure_eigen_vectors_number(
self.eigen_values, model_configuration.e_values,
model_configuration.x0_values,
disease_hypothesis.get_all_disease_indices())
if self.eigen_vectors_number == disease_hypothesis.number_of_regions:
# Calculate the propagation strength index by summing all eigenvectors
lsa_propagation_strength = numpy.abs(
numpy.sum(self.eigen_vectors, axis=1))
else:
sorted_indices = max(self.eigen_vectors_number, 1)
# Calculate the propagation strength index by summing the first n eigenvectors (minimum 1)
if self.weighted_eigenvector_sum:
lsa_propagation_strength = numpy.abs(
weighted_vector_sum(self.eigen_values[:sorted_indices],
self.eigen_vectors[:, :sorted_indices],
normalize=True))
else:
lsa_propagation_strength = numpy.abs(
numpy.sum(self.eigen_vectors[:, :sorted_indices], axis=1))
if self.normalize_propagation_strength:
# Normalize by the maximum
lsa_propagation_strength /= numpy.max(lsa_propagation_strength)
# # TODO: this has to be corrected
# if self.eigen_vectors_number < 0.2 * disease_hypothesis.number_of_regions:
# propagation_strength_elbow = numpy.max([self.get_curve_elbow_point(lsa_propagation_strength),
# self.eigen_vectors_number])
# else:
propagation_strength_elbow = self.get_curve_elbow_point(
lsa_propagation_strength)
propagation_indices = lsa_propagation_strength.argsort(
)[-propagation_strength_elbow:]
hypothesis_builder = HypothesisBuilder(disease_hypothesis.number_of_regions).\
set_attributes_based_on_hypothesis(disease_hypothesis). \
set_lsa_propagation(propagation_indices, lsa_propagation_strength)
return hypothesis_builder.build_lsa_hypothesis()
def run_lsa(self, disease_hypothesis, model_configuration):
if self.lsa_method == "auto":
if numpy.any(model_configuration.x1eq > X1EQ_CR_DEF):
self.lsa_method = "2D"
else:
self.lsa_method = "1D"
if self.lsa_method == "2D" and numpy.all(model_configuration.x1eq <= X1EQ_CR_DEF):
warning("LSA with the '2D' method (on the 2D Epileptor model) will not produce interpretable results when"
" the equilibrium point of the system is not supercritical (unstable)!")
jacobian = self._compute_jacobian(model_configuration)
# Perform eigenvalue decomposition
eigen_values, eigen_vectors = numpy.linalg.eig(jacobian)
eigen_values = numpy.real(eigen_values)
eigen_vectors = numpy.real(eigen_vectors)
sorted_indices = numpy.argsort(eigen_values, kind='mergesort')
if self.lsa_method == "2D":
sorted_indices = sorted_indices[::-1]
self.eigen_vectors = eigen_vectors[:, sorted_indices]
self.eigen_values = eigen_values[sorted_indices]
self._ensure_eigen_vectors_number(self.eigen_values[:disease_hypothesis.number_of_regions],
model_configuration.e_values, model_configuration.x0_values,
disease_hypothesis.regions_disease_indices)
if self.eigen_vectors_number == disease_hypothesis.number_of_regions:
# Calculate the propagation strength index by summing all eigenvectors
lsa_propagation_strength = numpy.abs(numpy.sum(self.eigen_vectors, axis=1))
else:
sorted_indices = max(self.eigen_vectors_number, 1)
# Calculate the propagation strength index by summing the first n eigenvectors (minimum 1)
if self.weighted_eigenvector_sum:
lsa_propagation_strength = \
numpy.abs(weighted_vector_sum(numpy.array(self.eigen_values[:self.eigen_vectors_number]),
numpy.array(self.eigen_vectors[:, :self.eigen_vectors_number]),
normalize=True))
else:
lsa_propagation_strength = \
numpy.abs(numpy.sum(self.eigen_vectors[:, :self.eigen_vectors_number], axis=1))
if self.lsa_method == "2D":
# lsa_propagation_strength = lsa_propagation_strength[:disease_hypothesis.number_of_regions]
# or
# lsa_propagation_strength = numpy.where(lsa_propagation_strength[:disease_hypothesis.number_of_regions] >=
# lsa_propagation_strength[disease_hypothesis.number_of_regions:],
# lsa_propagation_strength[:disease_hypothesis.number_of_regions],
# lsa_propagation_strength[disease_hypothesis.number_of_regions:])
# or
lsa_propagation_strength = numpy.sqrt(lsa_propagation_strength[:disease_hypothesis.number_of_regions]**2 +
lsa_propagation_strength[disease_hypothesis.number_of_regions:]**2)
lsa_propagation_strength = numpy.log10(lsa_propagation_strength)
lsa_propagation_strength -= lsa_propagation_strength.min()
if self.normalize_propagation_strength:
# Normalize by the maximum
lsa_propagation_strength /= numpy.max(lsa_propagation_strength)
# # TODO: this has to be corrected
# if self.eigen_vectors_number < 0.2 * disease_hypothesis.number_of_regions:
# propagation_strength_elbow = numpy.max([self.get_curve_elbow_point(lsa_propagation_strength),
# self.eigen_vectors_number])
# else:
propagation_strength_elbow = self.get_curve_elbow_point(lsa_propagation_strength)
propagation_indices = lsa_propagation_strength.argsort()[-propagation_strength_elbow:]
hypothesis_builder = HypothesisBuilder(disease_hypothesis.number_of_regions). \
set_attributes_based_on_hypothesis(disease_hypothesis). \
set_name(disease_hypothesis.name + "_LSA"). \
set_lsa_propagation(propagation_indices, lsa_propagation_strength)
return hypothesis_builder.build_lsa_hypothesis()