def __init__(self,
frequency,
wavelength,
flux,
noise,
voigt,
chi_limit=2.5):
self.frequency = frequency
self.wavelength = wavelength
self.flux = flux
self.noise = noise
self.num_pixels = len(frequency)
self.voigt = voigt
self.dataset = Spectrum(self.frequency, self.wavelength, self.flux,
self.noise)
"""
Make initial guess for number of local minima in the region.
Smooth the spectra with a gaussian and find the number of local minima.
as a safety precaucion, set the initial guess for number of profiles to 1 if
there are less than 4 local minima.
"""
self.initial_n = argrelextrema(gaussian_filter(self.dataset.flux, 3),
np.less)[0].shape[0]
if self.initial_n < 4:
self.initial_n = 1.
self.n = self.initial_n.copy()
def find_good_fit(mode, dataset: Spectrum):
n_initial = dataset.estimate_n()
n_components = n_initial + 0
phases = []
results = []
evidences = []
good_fit = False
evidences.insert(0, -1. * np.inf)
i = 1
while not good_fit:
phase = make_phase(mode, n_components)
result = phase.run(dataset=dataset)
evidence = result.output.evidence
# I think we need to just stop after a certain chi squared tbh
### Evaluate the result
if evidence > evidences[i - 1]:
n_components += 1
i += 1
phases.append(phase)
results.append(result)
evidences.append(evidence)
else:
good_fit == True
return phases, results, evidences
def new_region_spectra(self):
new_spectra = []
for reg in self.region_pixels:
new_spectra.append(
Spectrum(self.dataset.frequency[reg[0]:reg[1]],
self.dataset.wavelength[reg[0]:reg[1]],
self.dataset.flux[reg[0]:reg[1]],
self.dataset.noise[reg[0]:reg[1]]))
return new_spectra
# Setup the path to the config folder, using the autolens_workspace path.
config_path = workspace_path + "config"
# Use this path to explicitly set the config path and output path.
af.conf.instance = af.conf.Config(
config_path=config_path, output_path=workspace_path + "output"
)
from vamp_workspace.make_data import FakeGauss
fakeGaussA = FakeGauss(center=-1.0, sigma=2.0, intensity=0.5)
fakeGaussB = FakeGauss(center=1.5, sigma=1.0, intensity=1.0)
fakeGauss_2comp = fakeGaussB.gauss + fakeGaussA.gauss + fakeGaussA.noise
dataset = Spectrum(fakeGaussA.x, fakeGaussA.x, 1.0 - fakeGauss_2comp, fakeGaussA.noise)
phase = ph.Phase(
phase_name="phase_x2_gaussians",
profiles=af.CollectionPriorModel(
gaussian_0=profile_models.Gaussian, gaussian_1=profile_models.Gaussian
),
)
result = phase.run(dataset=dataset)
plt.plot(dataset.frequency, result.most_likely_model_spectrum)
plt.plot(dataset.frequency, dataset.flux)
plt.show()
# We also have an 'output' attribute, which in this case is a MultiNestOutput object:
print(result.output)
# Setup the path to the config folder, using the vamp_workspace path.
config_path = workspace_path + "config"
# Use this path to explicitly set the config path and output path.
af.conf.instance = af.conf.Config(config_path=config_path,
output_path=workspace_path + "output")
from vamp_workspace.make_data import FakeGauss
fakeGauss = FakeGauss(center=-1.0, sigma=2.0, intensity=0.5)
phase = ph.Phase(
phase_name="phase_x1_gaussians",
profiles=af.CollectionPriorModel(gaussian_0=profile_models.Gaussian))
dataset = Spectrum(fakeGauss.x, fakeGauss.x, 1.0 - fakeGauss.noisy_gauss,
fakeGauss.noise)
result = phase.run(dataset=dataset)
plt.plot(dataset.frequency, result.most_likely_model_spectrum)
plt.plot(dataset.frequency, dataset.flux)
plt.show()
# We also have an 'output' attribute, which in this case is a MultiNestOutput object:
print(result.output)
# This object acts as an interface between the MultiNest output results on your hard-disk and this Python code. For
# example, we can use it to get the evidence estimated by MultiNest.
print(result.output.evidence)
# We can also use it to get a model-instance of the "most probable" model, which is the model where each parameter is
# the value estimated from the probability distribution of parameter space.
mp_instance = result.output.most_probable_instance
print()
# Setup the path to the config folder, using the vamp_workspace path.
config_path = workspace_path + "config"
# Use this path to explicitly set the config path and output path.
af.conf.instance = af.conf.Config(config_path=config_path,
output_path=workspace_path + "output")
from vamp_workspace.make_data import FakeVoigt
fakeVoigt = FakeVoigt(center=-1.0, intensity=0.5, fwhm_L=1.0, fwhm_G=2.0)
phase = ph.Phase(
phase_name="phase_x1_voigt",
profiles=af.CollectionPriorModel(voigt_0=profile_models.Voigt))
dataset = Spectrum(fakeVoigt.x, fakeVoigt.x, 1.0 - fakeVoigt.noisy_voigt,
fakeVoigt.noise)
result = phase.run(dataset=dataset)
# We also have an 'output' attribute, which in this case is a MultiNestOutput object:
print(result.output)
# This object acts as an interface between the MultiNest output results on your hard-disk and this Python code. For
# example, we can use it to get the evidence estimated by MultiNest.
print(result.output.evidence)
# We can also use it to get a model-instance of the "most probable" model, which is the model where each parameter is
# the value estimated from the probability distribution of parameter space.
mp_instance = result.output.most_probable_instance
print()
print("Most Probable Model:\n")
print("Centre = ", [i.center for i in mp_instance.profiles])
print("Intensity = ", [i.intensity for i in mp_instance.profiles])
centers_c = [2.]
sigmas_a = [1.0]
sigmas_b = [1.5]
sigmas_c = [2.0]
intensity_a = [0.4]
intensity_b = [1.0]
intensity_c = [0.7]
combos = ['e']
for i, combo in enumerate(combos):
attrs = {}
fakeGaussA = FakeGauss(center=centers_a[i], sigma=sigmas_a[i], intensity=intensity_a[i])
fakeGaussB = FakeGauss(center=centers_b[i], sigma=sigmas_b[i], intensity=intensity_b[i])
fakeGaussC = FakeGauss(center=centers_c[i], sigma=sigmas_c[i], intensity=intensity_c[i])
dataset_a = Spectrum(fakeGaussA.x, fakeGaussA.x, 1.0 - fakeGaussA.gauss - fakeGaussA.noise, fakeGaussA.noise)
attrs['center_0'] = centers_a[i]; attrs['sigma_0'] = sigmas_a[i]; attrs['intensity_0'] = intensity_a[i]
dataset_a.save_as_h5py(spectrum_dir+'combo_' + combo + '_1_component.h5', attributes=attrs)
dataset_a.plot_spectrum(spectrum_dir+'combo_' + combo + '_1_component.png')
dataset_b = Spectrum(fakeGaussB.x, fakeGaussB.x, 1.0 - fakeGaussA.gauss - fakeGaussB.gauss - fakeGaussB.noise, fakeGaussB.noise)
attrs['center_1'] = centers_b[i]; attrs['sigma_1'] = sigmas_b[i]; attrs['intensity_1'] = intensity_b[i]
dataset_b.save_as_h5py(spectrum_dir+'combo_' + combo + '_2_component.h5', attributes=attrs)
dataset_b.plot_spectrum(spectrum_dir+'combo_' + combo + '_2_component.png')
dataset_c = Spectrum(fakeGaussC.x, fakeGaussC.x, 1.0 - fakeGaussA.gauss - fakeGaussB.gauss - fakeGaussC.gauss - fakeGaussC.noise, fakeGaussC.noise)
attrs['center_2'] = centers_c[i]; attrs['sigma_2'] = sigmas_c[i]; attrs['intensity_2'] = intensity_c[i]
dataset_c.save_as_h5py(spectrum_dir+'combo_' + combo + '_3_component.h5', attributes=attrs)
dataset_c.plot_spectrum(spectrum_dir+'combo_' + combo + '_3_component.png')