def fit_qso(QSO_im, psf_ave, psf_std=None, source_params=None,ps_param=None, background_rms=0.04, pix_sz = 0.168,
exp_time = 300., fix_n=None, image_plot = True, corner_plot=True, supersampling_factor = 2,
flux_ratio_plot=False, deep_seed = False, fixcenter = False, QSO_msk=None, QSO_std=None,
tag = None, no_MCMC= False, pltshow = 1, return_Chisq = False, dump_result = False, pso_diag=False):
'''
A quick fit for the QSO image with (so far) single sersice + one PSF. The input psf noise is optional.
Parameter
--------
QSO_im: An array of the QSO image.
psf_ave: The psf image.
psf_std: The psf noise, optional.
source_params: The prior for the source. Default is given. If [], means no Sersic light.
background_rms: default as 0.04
exp_time: default at 2400.
deep_seed: if Ture, more mcmc steps will be performed.
tag: The name tag for save the plot
Return
--------
Will output the fitted image (Set image_plot = True), the corner_plot and the flux_ratio_plot.
source_result, ps_result, image_ps, image_host
To do
--------
'''
# data specifics need to set up based on the data situation
background_rms = background_rms # background noise per pixel (Gaussian)
exp_time = exp_time # exposure time (arbitrary units, flux per pixel is in units #photons/exp_time unit)
numPix = len(QSO_im) # cutout pixel size
deltaPix = pix_sz
psf_type = 'PIXEL' # 'gaussian', 'pixel', 'NONE'
kernel = psf_ave
kwargs_numerics = {'supersampling_factor': supersampling_factor, 'supersampling_convolution': False}
if source_params is None:
# here are the options for the host galaxy fitting
fixed_source = []
kwargs_source_init = []
kwargs_source_sigma = []
kwargs_lower_source = []
kwargs_upper_source = []
if fix_n == None:
fixed_source.append({}) # we fix the Sersic index to n=1 (exponential)
kwargs_source_init.append({'R_sersic': 0.3, 'n_sersic': 2., 'e1': 0., 'e2': 0., 'center_x': 0., 'center_y': 0.})
kwargs_source_sigma.append({'n_sersic': 0.5, 'R_sersic': 0.5, 'e1': 0.1, 'e2': 0.1, 'center_x': 0.1, 'center_y': 0.1})
kwargs_lower_source.append({'e1': -0.5, 'e2': -0.5, 'R_sersic': 0.1, 'n_sersic': 0.3, 'center_x': -10, 'center_y': -10})
kwargs_upper_source.append({'e1': 0.5, 'e2': 0.5, 'R_sersic': 3., 'n_sersic': 7., 'center_x': 10, 'center_y': 10})
elif fix_n is not None:
fixed_source.append({'n_sersic': fix_n})
kwargs_source_init.append({'R_sersic': 0.3, 'n_sersic': fix_n, 'e1': 0., 'e2': 0., 'center_x': 0., 'center_y': 0.})
kwargs_source_sigma.append({'n_sersic': 0.001, 'R_sersic': 0.5, 'e1': 0.1, 'e2': 0.1, 'center_x': 0.1, 'center_y': 0.1})
kwargs_lower_source.append({'e1': -0.5, 'e2': -0.5, 'R_sersic': 0.1, 'n_sersic': fix_n, 'center_x': -10, 'center_y': -10})
kwargs_upper_source.append({'e1': 0.5, 'e2': 0.5, 'R_sersic': 3, 'n_sersic': fix_n, 'center_x': 10, 'center_y': 10})
source_params = [kwargs_source_init, kwargs_source_sigma, fixed_source, kwargs_lower_source, kwargs_upper_source]
else:
source_params = source_params
if ps_param is None:
center_x = 0.0
center_y = 0.0
point_amp = QSO_im.sum()/2.
fixed_ps = [{}]
kwargs_ps = [{'ra_image': [center_x], 'dec_image': [center_y], 'point_amp': [point_amp]}]
kwargs_ps_init = kwargs_ps
kwargs_ps_sigma = [{'ra_image': [0.05], 'dec_image': [0.05]}]
kwargs_lower_ps = [{'ra_image': [-0.6], 'dec_image': [-0.6]}]
kwargs_upper_ps = [{'ra_image': [0.6], 'dec_image': [0.6]}]
ps_param = [kwargs_ps_init, kwargs_ps_sigma, fixed_ps, kwargs_lower_ps, kwargs_upper_ps]
else:
ps_param = ps_param
#==============================================================================
#Doing the QSO fitting
#==============================================================================
kwargs_data = sim_util.data_configure_simple(numPix, deltaPix, exp_time, background_rms, inverse=True)
data_class = ImageData(**kwargs_data)
kwargs_psf = {'psf_type': psf_type, 'kernel_point_source': kernel}
psf_class = PSF(**kwargs_psf)
data_class.update_data(QSO_im)
point_source_list = ['UNLENSED'] * len(ps_param[0])
pointSource = PointSource(point_source_type_list=point_source_list)
if fixcenter == False:
kwargs_constraints = {'num_point_source_list': [1] * len(ps_param[0])
}
elif fixcenter == True:
kwargs_constraints = {'joint_source_with_point_source': [[i, i] for i in range(len(ps_param[0]))],
'num_point_source_list': [1] * len(ps_param[0])
}
if source_params == []: #fitting image as Point source only.
kwargs_params = {'point_source_model': ps_param}
lightModel = None
kwargs_model = {'point_source_model_list': point_source_list }
imageModel = ImageModel(data_class, psf_class, point_source_class=pointSource, kwargs_numerics=kwargs_numerics)
kwargs_likelihood = {'check_bounds': True, #Set the bonds, if exceed, reutrn "penalty"
'image_likelihood_mask_list': [QSO_msk]
}
elif source_params != []:
kwargs_params = {'source_model': source_params,
'point_source_model': ps_param}
light_model_list = ['SERSIC_ELLIPSE'] * len(source_params[0])
lightModel = LightModel(light_model_list=light_model_list)
kwargs_model = { 'source_light_model_list': light_model_list,
'point_source_model_list': point_source_list
}
imageModel = ImageModel(data_class, psf_class, source_model_class=lightModel,
point_source_class=pointSource, kwargs_numerics=kwargs_numerics)
# numerical options and fitting sequences
kwargs_likelihood = {'check_bounds': True, #Set the bonds, if exceed, reutrn "penalty"
'source_marg': False, #In likelihood_module.LikelihoodModule -- whether to fully invert the covariance matrix for marginalization
'check_positive_flux': True,
'image_likelihood_mask_list': [QSO_msk]
}
kwargs_data['image_data'] = QSO_im
if QSO_std is not None:
kwargs_data['noise_map'] = QSO_std
if psf_std is not None:
kwargs_psf['psf_error_map'] = psf_std
image_band = [kwargs_data, kwargs_psf, kwargs_numerics]
multi_band_list = [image_band]
kwargs_data_joint = {'multi_band_list': multi_band_list, 'multi_band_type': 'multi-linear'} # 'single-band', 'multi-linear', 'joint-linear'
fitting_seq = FittingSequence(kwargs_data_joint, kwargs_model, kwargs_constraints, kwargs_likelihood, kwargs_params)
if deep_seed == False:
fitting_kwargs_list = [
['PSO', {'sigma_scale': 0.8, 'n_particles': 100, 'n_iterations': 60}],
['MCMC', {'n_burn': 10, 'n_run': 10, 'walkerRatio': 50, 'sigma_scale': .1}]
]
elif deep_seed == True:
fitting_kwargs_list = [
['PSO', {'sigma_scale': 0.8, 'n_particles': 250, 'n_iterations': 250}],
['MCMC', {'n_burn': 100, 'n_run': 200, 'walkerRatio': 10, 'sigma_scale': .1}]
]
if no_MCMC == True:
fitting_kwargs_list = [fitting_kwargs_list[0],
]
start_time = time.time()
chain_list = fitting_seq.fit_sequence(fitting_kwargs_list)
kwargs_result = fitting_seq.best_fit()
ps_result = kwargs_result['kwargs_ps']
source_result = kwargs_result['kwargs_source']
if no_MCMC == False:
sampler_type, samples_mcmc, param_mcmc, dist_mcmc = chain_list[1]
end_time = time.time()
print(end_time - start_time, 'total time needed for computation')
print('============ CONGRATULATION, YOUR JOB WAS SUCCESSFUL ================ ')
imageLinearFit = ImageLinearFit(data_class=data_class, psf_class=psf_class,
source_model_class=lightModel,
point_source_class=pointSource,
kwargs_numerics=kwargs_numerics)
image_reconstructed, error_map, _, _ = imageLinearFit.image_linear_solve(kwargs_source=source_result, kwargs_ps=ps_result)
# this is the linear inversion. The kwargs will be updated afterwards
modelPlot = ModelPlot(multi_band_list, kwargs_model, kwargs_result,
arrow_size=0.02, cmap_string="gist_heat", likelihood_mask_list=[QSO_msk])
image_host = [] #!!! The linear_solver before and after LensModelPlot could have different result for very faint sources.
for i in range(len(source_result)):
image_host.append(imageModel.source_surface_brightness(source_result, de_lensed=True,unconvolved=False,k=i))
image_ps = []
for i in range(len(ps_result)):
image_ps.append(imageModel.point_source(ps_result, k = i))
if pso_diag == True:
f, axes = chain_plot.plot_chain_list(chain_list,0)
if pltshow == 0:
plt.close()
else:
plt.show()
# let's plot the output of the PSO minimizer
reduced_Chisq = imageLinearFit.reduced_chi2(image_reconstructed, error_map)
if image_plot:
f, axes = plt.subplots(3, 3, figsize=(16, 16), sharex=False, sharey=False)
modelPlot.data_plot(ax=axes[0,0], text="Data")
modelPlot.model_plot(ax=axes[0,1])
modelPlot.normalized_residual_plot(ax=axes[0,2], v_min=-6, v_max=6)
modelPlot.decomposition_plot(ax=axes[1,0], text='Host galaxy', source_add=True, unconvolved=True)
modelPlot.decomposition_plot(ax=axes[1,1], text='Host galaxy convolved', source_add=True)
modelPlot.decomposition_plot(ax=axes[1,2], text='All components convolved', source_add=True, lens_light_add=True, point_source_add=True)
modelPlot.subtract_from_data_plot(ax=axes[2,0], text='Data - Point Source', point_source_add=True)
modelPlot.subtract_from_data_plot(ax=axes[2,1], text='Data - host galaxy', source_add=True)
modelPlot.subtract_from_data_plot(ax=axes[2,2], text='Data - host galaxy - Point Source', source_add=True, point_source_add=True)
f.tight_layout()
#f.subplots_adjust(left=None, bottom=None, right=None, top=None, wspace=0., hspace=0.05)
if tag is not None:
f.savefig('{0}_fitted_image.pdf'.format(tag))
if pltshow == 0:
plt.close()
else:
plt.show()
if corner_plot==True and no_MCMC==False:
# here the (non-converged) MCMC chain of the non-linear parameters
if not samples_mcmc == []:
n, num_param = np.shape(samples_mcmc)
plot = corner.corner(samples_mcmc, labels=param_mcmc, show_titles=True)
if tag is not None:
plot.savefig('{0}_para_corner.pdf'.format(tag))
plt.close()
# if pltshow == 0:
# plt.close()
# else:
# plt.show()
if flux_ratio_plot==True and no_MCMC==False:
param = Param(kwargs_model, kwargs_fixed_source=source_params[2], kwargs_fixed_ps=ps_param[2], **kwargs_constraints)
mcmc_new_list = []
if len(ps_param[2]) == 1:
labels_new = ["Quasar flux"] + ["host{0} flux".format(i) for i in range(len(source_params[0]))]
else:
labels_new = ["Quasar{0} flux".format(i) for i in range(len(ps_param[2]))] + ["host{0} flux".format(i) for i in range(len(source_params[0]))]
if len(samples_mcmc) > 10000:
trans_steps = [len(samples_mcmc)-10000, len(samples_mcmc)]
else:
trans_steps = [0, len(samples_mcmc)]
for i in range(trans_steps[0], trans_steps[1]):
kwargs_out = param.args2kwargs(samples_mcmc[i])
kwargs_light_source_out = kwargs_out['kwargs_source']
kwargs_ps_out = kwargs_out['kwargs_ps']
image_reconstructed, _, _, _ = imageLinearFit.image_linear_solve(kwargs_source=kwargs_light_source_out, kwargs_ps=kwargs_ps_out)
flux_quasar = []
if len(ps_param[0]) == 1:
image_ps_j = imageModel.point_source(kwargs_ps_out)
flux_quasar.append(np.sum(image_ps_j))
else:
for j in range(len(ps_param[0])):
image_ps_j = imageModel.point_source(kwargs_ps_out, k=j)
flux_quasar.append(np.sum(image_ps_j))
fluxs = []
for j in range(len(source_params[0])):
image_j = imageModel.source_surface_brightness(kwargs_light_source_out,unconvolved= False, k=j)
fluxs.append(np.sum(image_j))
mcmc_new_list.append(flux_quasar + fluxs )
if int(i/1000) > int((i-1)/1000) :
print(len(samples_mcmc), "MCMC samplers in total, finished translate:", i )
plot = corner.corner(mcmc_new_list, labels=labels_new, show_titles=True)
if tag is not None:
plot.savefig('{0}_HOSTvsQSO_corner.pdf'.format(tag))
if pltshow == 0:
plt.close()
else:
plt.show()
if QSO_std is None:
noise_map = np.sqrt(data_class.C_D+np.abs(error_map))
else:
noise_map = np.sqrt(QSO_std**2+np.abs(error_map))
if dump_result == True:
if flux_ratio_plot==True and no_MCMC==False:
trans_paras = [mcmc_new_list, labels_new, 'mcmc_new_list, labels_new']
else:
trans_paras = []
picklename= tag + '.pkl'
best_fit = [source_result, image_host, ps_result, image_ps,'source_result, image_host, ps_result, image_ps']
chain_list_result = [chain_list, 'chain_list']
kwargs_fixed_source=source_params[2]
kwargs_fixed_ps=ps_param[2]
classes = data_class, psf_class, lightModel, pointSource
material = multi_band_list, kwargs_model, kwargs_result, QSO_msk, kwargs_fixed_source, kwargs_fixed_ps, kwargs_constraints, kwargs_numerics, classes
pickle.dump([best_fit, chain_list_result, trans_paras, material], open(picklename, 'wb'))
if return_Chisq == False:
return source_result, ps_result, image_ps, image_host, noise_map
elif return_Chisq == True:
return source_result, ps_result, image_ps, image_host, noise_map, reduced_Chisq
pointSource = PointSource(point_source_type_list=point_source_list)
kwargs_numerics = {'subgrid_res': 1, 'psf_subgrid': False}
#kwargs_numerics['mask'] = QSO_msk
imageModel = ImageModel(data_class,
psf_class,
source_model_class=lightModel,
point_source_class=pointSource,
kwargs_numerics=kwargs_numerics)
#labels_new = ["Quasar flux"] + ["host{0} flux".format(i) for i in range(3)] + ["host{0} Reff".format(i) for i in range(3)]
for i in range(len(samples_mcmc) / 10):
kwargs_lens_out, kwargs_light_source_out, kwargs_light_lens_out, kwargs_ps_out, kwargs_cosmo = param.args2kwargs(
samples_mcmc[i + len(samples_mcmc) / 10 * 9])
image_reconstructed, _, _, _ = imageModel.image_linear_solve(
kwargs_source=kwargs_light_source_out, kwargs_ps=kwargs_ps_out)
image_ps = imageModel.point_source(kwargs_ps_out)
flux_quasar = np.sum(image_ps)
fluxs, reffs = [], []
for j in range(3):
image_j = imageModel.source_surface_brightness(kwargs_light_source_out,
unconvolved=False,
k=j)
fluxs.append(np.sum(image_j))
reffs.append(kwargs_light_source_out[j]['R_sersic'])
mcmc_new_list.append([flux_quasar] + fluxs + reffs)
if i / 1000 > (i - 1) / 1000:
print "finished translate:", i
print 'The new translated plot:'
plot = corner.corner(mcmc_new_list, labels=labels_new, show_titles=True)
plt.show()
def sim_image(self, info_dict):
"""
Simulate an image based on specifications in sim_dict
Args:
info_dict (dict): A single element from the list produced interanlly by input_reader.Organizer.breakup().
Contains all the properties of a single image to generate.
"""
output_image = []
if self.return_planes:
output_source, output_lens, output_point_source, output_noise = [], [], [], []
output_metadata = []
#set the cosmology
cosmology_info = ['H0', 'Om0', 'Tcmb0', 'Neff', 'm_nu', 'Ob0']
cosmo = FlatLambdaCDM(
**dict_select_choose(list(info_dict.values())[0], cosmology_info))
for band, sim_dict in info_dict.items():
# Parse the info dict
params = self.parse_single_band_info_dict(sim_dict,
cosmo,
band=band)
kwargs_single_band = params[0]
kwargs_model = params[1]
kwargs_numerics = params[2]
kwargs_lens_light_list = params[3]
kwargs_source_list = params[4]
kwargs_point_source_list = params[5]
kwargs_lens_model_list = params[6]
output_metadata += params[7]
# Make image
# data properties
kwargs_data = sim_util.data_configure_simple(
sim_dict['numPix'], kwargs_single_band['pixel_scale'],
kwargs_single_band['exposure_time'])
data_class = ImageData(**kwargs_data)
# psf properties
kwargs_psf = {
'psf_type': kwargs_single_band['psf_type'],
'pixel_size': kwargs_single_band['pixel_scale'],
'fwhm': kwargs_single_band['seeing']
}
psf_class = PSF(**kwargs_psf)
# SimAPI instance for conversion to observed quantities
sim = SimAPI(numpix=sim_dict['numPix'],
kwargs_single_band=kwargs_single_band,
kwargs_model=kwargs_model)
kwargs_lens_model_list = sim.physical2lensing_conversion(
kwargs_mass=kwargs_lens_model_list)
kwargs_lens_light_list, kwargs_source_list, _ = sim.magnitude2amplitude(
kwargs_lens_light_mag=kwargs_lens_light_list,
kwargs_source_mag=kwargs_source_list)
# lens model properties
lens_model_class = LensModel(
lens_model_list=kwargs_model['lens_model_list'],
z_lens=kwargs_model['lens_redshift_list'][0],
z_source=kwargs_model['z_source'],
cosmo=cosmo)
# source properties
source_model_class = LightModel(
light_model_list=kwargs_model['source_light_model_list'])
# lens light properties
lens_light_model_class = LightModel(
light_model_list=kwargs_model['lens_light_model_list'])
# solve for PS positions to incorporate time delays
lensEquationSolver = LensEquationSolver(lens_model_class)
kwargs_ps = []
for ps_idx, ps_mag in enumerate(kwargs_point_source_list):
# modify the SimAPI instance to do one point source at a time
temp_kwargs_model = {k: v for k, v in kwargs_model.items()}
temp_kwargs_model['point_source_model_list'] = [
kwargs_model['point_source_model_list'][ps_idx]
]
sim = SimAPI(numpix=sim_dict['numPix'],
kwargs_single_band=kwargs_single_band,
kwargs_model=temp_kwargs_model)
if kwargs_model['point_source_model_list'][
ps_idx] == 'SOURCE_POSITION':
# convert each image to an amplitude
amplitudes = []
for mag in ps_mag['magnitude']:
ps_dict = {k: v for k, v in ps_mag.items()}
ps_dict['magnitude'] = mag
_, _2, ps = sim.magnitude2amplitude(
kwargs_ps_mag=[ps_dict])
amplitudes.append(ps[0]['source_amp'])
x_image, y_image = lensEquationSolver.findBrightImage(
ps[0]['ra_source'],
ps[0]['dec_source'],
kwargs_lens_model_list,
numImages=4, # max number of images
min_distance=kwargs_single_band['pixel_scale'],
search_window=sim_dict['numPix'] *
kwargs_single_band['pixel_scale'])
magnification = lens_model_class.magnification(
x_image, y_image, kwargs=kwargs_lens_model_list)
#amplitudes = np.array(amplitudes) * np.abs(magnification)
amplitudes = np.array(
[a * m for a, m in zip(amplitudes, magnification)])
kwargs_ps.append({
'ra_image': x_image,
'dec_image': y_image,
'point_amp': amplitudes
})
else:
_, _2, ps = sim.magnitude2amplitude(kwargs_ps_mag=[ps_mag])
kwargs_ps.append(ps[0])
# point source properties
point_source_class = PointSource(point_source_type_list=[
x if x != 'SOURCE_POSITION' else 'LENSED_POSITION'
for x in kwargs_model['point_source_model_list']
],
fixed_magnification_list=[False] *
len(kwargs_ps))
# create an image model
image_model = ImageModel(data_class,
psf_class,
lens_model_class,
source_model_class,
lens_light_model_class,
point_source_class,
kwargs_numerics=kwargs_numerics)
# generate image
image_sim = image_model.image(kwargs_lens_model_list,
kwargs_source_list,
kwargs_lens_light_list, kwargs_ps)
poisson = image_util.add_poisson(
image_sim, exp_time=kwargs_single_band['exposure_time'])
sigma_bkg = data_util.bkg_noise(
kwargs_single_band['read_noise'],
kwargs_single_band['exposure_time'],
kwargs_single_band['sky_brightness'],
kwargs_single_band['pixel_scale'],
num_exposures=kwargs_single_band['num_exposures'])
bkg = image_util.add_background(image_sim, sigma_bkd=sigma_bkg)
image = image_sim + bkg + poisson
# Save theta_E (and sigma_v if used)
for ii in range(len(output_metadata)):
output_metadata.append({
'PARAM_NAME':
output_metadata[ii]['PARAM_NAME'].replace(
'sigma_v', 'theta_E'),
'PARAM_VALUE':
kwargs_lens_model_list[output_metadata[ii]
['LENS_MODEL_IDX']]['theta_E'],
'LENS_MODEL_IDX':
output_metadata[ii]['LENS_MODEL_IDX']
})
# Solve lens equation if desired
if self.solve_lens_equation:
#solver = lens_equation_solver.LensEquationSolver(imSim.LensModel)
#x_mins, y_mins = solver.image_position_from_source(sourcePos_x=kwargs_source_list[0]['center_x'],
# sourcePos_y=kwargs_source_list[0]['center_y'],
# kwargs_lens=kwargs_lens_model_list)
x_mins, y_mins = x_image, y_image
num_source_images = len(x_mins)
# Add noise
image_noise = np.zeros(np.shape(image))
for noise_source_num in range(
1, sim_dict['NUMBER_OF_NOISE_SOURCES'] + 1):
image_noise += self._generate_noise(
sim_dict['NOISE_SOURCE_{0}-NAME'.format(noise_source_num)],
np.shape(image),
select_params(
sim_dict,
'NOISE_SOURCE_{0}-'.format(noise_source_num)))
image += image_noise
# Combine with other bands
output_image.append(image)
# Store plane-separated info if requested
if self.return_planes:
output_lens.append(
image_model.lens_surface_brightness(
kwargs_lens_light_list))
output_source.append(
image_model.source_surface_brightness(
kwargs_source_list, kwargs_lens_model_list))
output_point_source.append(
image_model.point_source(kwargs_ps,
kwargs_lens_model_list))
output_noise.append(image_noise)
# Return the desired information in a dictionary
return_dict = {
'output_image': np.array(output_image),
'output_lens_plane': None,
'output_source_plane': None,
'output_point_source_plane': None,
'output_noise_plane': None,
'x_mins': None,
'y_mins': None,
'num_source_images': None,
'additional_metadata': output_metadata
}
if self.return_planes:
return_dict['output_lens_plane'] = np.array(output_lens)
return_dict['output_source_plane'] = np.array(output_source)
return_dict['output_point_source_plane'] = np.array(
output_point_source)
return_dict['output_noise_plane'] = np.array(output_noise)
if self.solve_lens_equation:
return_dict['x_mins'] = x_mins
return_dict['y_mins'] = y_mins
return_dict['num_source_images'] = num_source_images
return return_dict
