def test_multilengthdistribution_plot(
list_of_length_arrays,
list_of_area_values,
names,
automatic_cut_offs,
using_branches,
num_regression,
):
"""
Test MultiLengthDistribution plot_multi_length_distributions.
"""
assert isinstance(list_of_length_arrays, list)
assert isinstance(list_of_area_values, list)
assert isinstance(names, list)
assert len(list_of_length_arrays) == len(list_of_area_values) == len(names)
multi_length_distribution = length_distributions.MultiLengthDistribution(
distributions=[
length_distributions.LengthDistribution(
name=name,
lengths=lengths,
area_value=area_value,
using_branches=using_branches,
)
for name, lengths, area_value in zip(
names, list_of_length_arrays, list_of_area_values
)
],
using_branches=using_branches,
)
polyfit, fig, ax = multi_length_distribution.plot_multi_length_distributions(
automatic_cut_offs=automatic_cut_offs, plot_truncated_data=True
)
plt.close()
mld = multi_length_distribution
(
truncated_length_array_all,
ccm_array_normed_all,
_,
_,
) = mld.normalized_distributions(
automatic_cut_offs=automatic_cut_offs,
)
assert isinstance(fig, Figure) and isinstance(ax, Axes)
assert isinstance(truncated_length_array_all, list)
assert isinstance(ccm_array_normed_all, list)
num_regression.check(
{
"concatted_lengths": np.concatenate(truncated_length_array_all),
"concatted_ccm": np.concatenate(ccm_array_normed_all),
"fitted_y_values": polyfit.y_fit,
"m_value": np.array([polyfit.m_value]),
"constant": np.array([polyfit.constant]),
},
default_tolerance=dict(atol=1e-4, rtol=1e-4),
)
def test_multilengthdistribution_plot(
list_of_length_arrays,
list_of_area_values,
names,
cut_distributions,
using_branches,
num_regression,
):
"""
Test MultiLengthDistribution plot_multi_length_distributions.
"""
assert isinstance(list_of_length_arrays, list)
assert isinstance(list_of_area_values, list)
assert isinstance(names, list)
assert len(list_of_length_arrays) == len(list_of_area_values) == len(names)
multi_length_distribution = length_distributions.MultiLengthDistribution(
distributions=[
length_distributions.LengthDistribution(
name=name,
lengths=lengths,
area_value=area_value,
using_branches=using_branches,
) for name, lengths, area_value in zip(
names, list_of_length_arrays, list_of_area_values)
],
using_branches=using_branches,
cut_distributions=cut_distributions,
)
fig, ax = multi_length_distribution.plot_multi_length_distributions()
mld = multi_length_distribution
assert isinstance(fig, Figure) and isinstance(ax, Axes)
assert isinstance(mld.truncated_length_array_all, list)
assert isinstance(mld.ccm_array_normed_all, list)
assert isinstance(mld.fitted_y_values, np.ndarray)
assert isinstance(mld.m_value, float)
assert isinstance(mld.constant, float)
assert isinstance(mld.concatted_lengths, np.ndarray)
assert isinstance(mld.concatted_ccm, np.ndarray)
num_regression.check(
{
"concatted_lengths": mld.concatted_lengths,
"concatted_ccm": mld.concatted_ccm,
"fitted_y_values": mld.fitted_y_values,
"m_value": np.array([mld.m_value]),
"constant": np.array([mld.constant]),
},
default_tolerance=dict(atol=1e-4, rtol=1e-4),
)
def multi_length_distributions(
self, using_branches: bool
) -> length_distributions.MultiLengthDistribution:
"""
Get MultiLengthDistribution of all networks.
"""
distributions_dict = self.network_length_distributions(
using_branches=using_branches, using_azimuth_sets=False)
distributions = [
distributions_dict[network.name][network.name]
for network in self.networks
]
multi_distribution = length_distributions.MultiLengthDistribution(
distributions=distributions,
using_branches=using_branches,
)
return multi_distribution
def test_plot_mld_optimized(distributions):
"""
Test plot_mld_optimized.
"""
mld = length_distributions.MultiLengthDistribution(
distributions=distributions, using_branches=False
)
opt_result, opt_mld = mld.optimize_cut_offs()
polyfit, fig, ax = opt_mld.plot_multi_length_distributions(
automatic_cut_offs=False, plot_truncated_data=True
)
assert isinstance(opt_result, length_distributions.MultiScaleOptimizationResult)
assert isinstance(polyfit, length_distributions.Polyfit)
assert isinstance(fig, Figure) and isinstance(ax, Axes)
plt.close()
def set_multi_length_distributions(
self, using_branches: bool
) -> Dict[str, length_distributions.MultiLengthDistribution]:
"""
Get set-wise multi-length distributions.
"""
distributions_dict = self.network_length_distributions(
using_branches=using_branches, using_azimuth_sets=True)
mlds = dict()
for set_name in self.networks[0].azimuth_set_names:
set_lengths: List[length_distributions.LengthDistribution] = []
for lds in distributions_dict.values():
set_lengths.append(lds[set_name])
mld = length_distributions.MultiLengthDistribution(
distributions=set_lengths,
using_branches=using_branches,
)
mlds[set_name] = mld
return mlds
def test_fit_to_multi_scale_lengths_fitter_comparisons(
distributions: List[length_distributions.LengthDistribution],
automatic_cut_offs: bool,
):
"""
Test fit_to_multi_scale_lengths.
"""
mld = length_distributions.MultiLengthDistribution(
distributions=distributions, using_branches=False
)
(
truncated_length_array_all,
ccm_array_normed_all,
# All length and ccm data ignored
_,
_,
) = mld.normalized_distributions(automatic_cut_offs=automatic_cut_offs)
concatted_lengths, concatted_ccm = np.concatenate(
truncated_length_array_all
), np.concatenate(ccm_array_normed_all)
# Fit with np.polyfit
numpy_polyfit_vals = length_distributions.fit_to_multi_scale_lengths(
lengths=concatted_lengths,
ccm=concatted_ccm,
fitter=length_distributions.numpy_polyfit,
)
# Fit with scikit LinearRegression
linear_regression_vals = length_distributions.fit_to_multi_scale_lengths(
lengths=concatted_lengths,
ccm=concatted_ccm,
fitter=length_distributions.scikit_linear_regression,
)
for val in [*numpy_polyfit_vals, *linear_regression_vals]:
# Polyfit attributes should be floats, array and callables (scorer func)
assert isinstance(val, (np.ndarray, float)) or callable(val)
def test_fit_to_multi_scale_lengths_fitter_comparisons(
distributions: List[length_distributions.LengthDistribution], ):
"""
Test fit_to_multi_scale_lengths.
"""
mld = length_distributions.MultiLengthDistribution(
distributions=distributions,
cut_distributions=True,
using_branches=False)
numpy_polyfit_vals = length_distributions.fit_to_multi_scale_lengths(
lengths=mld.concatted_lengths,
ccm=mld.concatted_ccm,
fitter=length_distributions.numpy_polyfit,
)
linear_regression_vals = length_distributions.fit_to_multi_scale_lengths(
lengths=mld.concatted_lengths,
ccm=mld.concatted_ccm,
fitter=length_distributions.scikit_linear_regression,
)
for val in [*numpy_polyfit_vals, *linear_regression_vals]:
assert isinstance(val, (np.ndarray, float))
mpl.rcParams["figure.figsize"] = (5, 5)
mpl.rcParams["font.size"] = 8
# %%
# Collect LengthDistributions into MultiLengthDistribution
# ------------------------------------------------------------------
networks = [kb11_network, hastholmen_network, lidar_200k_network]
distributions = [
netw.trace_length_distribution(azimuth_set=None) for netw in networks
]
mld = length_distributions.MultiLengthDistribution(
distributions=distributions,
using_branches=False,
fitter=length_distributions.scikit_linear_regression,
)
# %%
# Use scipy.optimize.shgo to optimize cut-off values
# ------------------------------------------------------------------
# See https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.shgo.html
# for potential keyword arguments to pass to the shgo call
# shgo_kwargs are passed as is to ``scipy.optimize.shgo``.
shgo_kwargs = dict(sampling_method="sobol", )
# Choose loss function for optimization. Here r2 is chosen to get a visually
# sensible result but it is generally ill-suited for optimizing cut-offs of
# power-law distributions.