def _run_pso(self,
n_particles,
n_iterations,
kwargs_fixed_lens,
kwargs_mean_lens,
kwargs_sigma_lens,
kwargs_fixed_source,
kwargs_mean_source,
kwargs_sigma_source,
kwargs_fixed_lens_light,
kwargs_mean_lens_light,
kwargs_sigma_lens_light,
kwargs_fixed_ps,
kwargs_mean_ps,
kwargs_sigma_ps,
kwargs_fixed_cosmo,
kwargs_mean_cosmo,
kwargs_sigma_cosmo,
threadCount=1,
mpi=False,
print_key='PSO',
sigma_factor=1,
compute_bool=None,
fix_solver=False):
# initialise mcmc classes
param_class = Param(self.kwargs_model,
self.kwargs_constraints,
kwargs_fixed_lens,
kwargs_fixed_source,
kwargs_fixed_lens_light,
kwargs_fixed_ps,
kwargs_fixed_cosmo,
self._lens_lower,
self._source_lower,
self._lens_light_lower,
self._ps_lower,
self._cosmo_lower,
self._lens_upper,
self._source_upper,
self._lens_light_upper,
self._ps_upper,
self._cosmo_upper,
kwargs_lens_init=kwargs_mean_lens,
fix_lens_solver=fix_solver)
init_pos = param_class.setParams(kwargs_mean_lens, kwargs_mean_source,
kwargs_mean_lens_light,
kwargs_mean_ps, kwargs_mean_cosmo)
sigma_start = param_class.setParams(kwargs_sigma_lens,
kwargs_sigma_source,
kwargs_sigma_lens_light,
kwargs_sigma_ps,
kwargs_sigma_cosmo)
lowerLimit = np.array(init_pos) - np.array(sigma_start) * sigma_factor
upperLimit = np.array(init_pos) + np.array(sigma_start) * sigma_factor
num_param, param_list = param_class.num_param()
# initialize ImSim() class
kwargs_likelihood = copy.deepcopy(self.kwargs_likelihood)
if compute_bool is not None:
kwargs_likelihood['bands_compute'] = compute_bool
imSim_class = class_creator.create_multiband(self.multi_band_list,
self.kwargs_model)
likelihoodModule = LikelihoodModule(
imSim_class=imSim_class,
param_class=param_class,
kwargs_likelihood=kwargs_likelihood)
# run PSO
mcmc_class = Sampler(likelihoodModule=likelihoodModule)
result, chain = mcmc_class.pso(n_particles,
n_iterations,
lowerLimit,
upperLimit,
init_pos=init_pos,
threadCount=threadCount,
mpi=mpi,
print_key=print_key)
lens_result, source_result, lens_light_result, ps_result, cosmo_result = param_class.getParams(
result, bijective=True)
return lens_result, source_result, lens_light_result, ps_result, cosmo_result, chain, param_list
def fit_qso_multiband(QSO_im_list, psf_ave_list, psf_std_list=None, source_params=None,ps_param=None,
background_rms_list=[0.04]*5, pix_sz = 0.168,
exp_time = 300., fix_n=None, image_plot = True, corner_plot=True,
flux_ratio_plot=True, deep_seed = False, fixcenter = False, QSO_msk_list=None,
QSO_std_list=None, tag = None, no_MCMC= False, pltshow = 1, new_band_seq=None):
'''
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.
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_list = background_rms_list # 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_list[0]) # cutout pixel size
deltaPix = pix_sz
psf_type = 'PIXEL' # 'gaussian', 'pixel', 'NONE'
kernel_list = psf_ave_list
if new_band_seq == None:
new_band_seq= range(len(QSO_im_list))
# if psf_std_list is not None:
# kwargs_numerics_list = [{'subgrid_res': 1, 'psf_subgrid': False, 'psf_error_map': True}] * len(QSO_im_list) #Turn on the PSF error map
# else:
kwargs_numerics_list = [{'supersampling_factor': 1, 'supersampling_convolution': False}] * len(QSO_im_list)
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 = []
# Disk component, as modelled by an elliptical Sersic profile
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_list[0].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': [0.01], 'dec_image': [0.01]}]
kwargs_lower_ps = [{'ra_image': [-10], 'dec_image': [-10]}]
kwargs_upper_ps = [{'ra_image': [10], 'dec_image': [10]}]
ps_param = [kwargs_ps_init, kwargs_ps_sigma, fixed_ps, kwargs_lower_ps, kwargs_upper_ps]
else:
ps_param = ps_param
kwargs_params = {'source_model': source_params,
'point_source_model': ps_param}
#==============================================================================
#Doing the QSO fitting
#==============================================================================
kwargs_data_list, data_class_list = [], []
for i in range(len(QSO_im_list)):
kwargs_data_i = sim_util.data_configure_simple(numPix, deltaPix, exp_time, background_rms_list[i], inverse=True)
kwargs_data_list.append(kwargs_data_i)
data_class_list.append(ImageData(**kwargs_data_i))
kwargs_psf_list = []
psf_class_list = []
for i in range(len(QSO_im_list)):
kwargs_psf_i = {'psf_type': psf_type, 'kernel_point_source': kernel_list[i]}
kwargs_psf_list.append(kwargs_psf_i)
psf_class_list.append(PSF(**kwargs_psf_i))
data_class_list[i].update_data(QSO_im_list[i])
light_model_list = ['SERSIC_ELLIPSE'] * len(source_params[0])
lightModel = LightModel(light_model_list=light_model_list)
point_source_list = ['UNLENSED']
pointSource = PointSource(point_source_type_list=point_source_list)
imageModel_list = []
for i in range(len(QSO_im_list)):
kwargs_data_list[i]['image_data'] = QSO_im_list[i]
# if QSO_msk_list is not None:
# kwargs_numerics_list[i]['mask'] = QSO_msk_list[i]
if QSO_std_list is not None:
kwargs_data_list[i]['noise_map'] = QSO_std_list[i]
# if psf_std_list is not None:
# kwargs_psf_list[i]['psf_error_map'] = psf_std_list[i]
image_band_list = []
for i in range(len(QSO_im_list)):
imageModel_list.append(ImageModel(data_class_list[i], psf_class_list[i], source_model_class=lightModel,
point_source_class=pointSource, kwargs_numerics=kwargs_numerics_list[i]))
image_band_list.append([kwargs_data_list[i], kwargs_psf_list[i], kwargs_numerics_list[i]])
multi_band_list = [image_band_list[i] for i in range(len(QSO_im_list))]
# numerical options and fitting sequences
kwargs_model = { 'source_light_model_list': light_model_list,
'point_source_model_list': point_source_list
}
if fixcenter == False:
kwargs_constraints = {'num_point_source_list': [1]
}
elif fixcenter == True:
kwargs_constraints = {'joint_source_with_point_source': [[0, 0]],
'num_point_source_list': [1]
}
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_list]
}
# mpi = False # MPI possible, but not supported through that notebook.
# The Params for the fitting. kwargs_init: initial input. kwargs_sigma: The parameter uncertainty. kwargs_fixed: fixed parameters;
#kwargs_lower,kwargs_upper: Lower and upper limits.
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': 80, 'n_iterations': 60, 'compute_bands': [True]+[False]*(len(QSO_im_list)-1)}],
['align_images', {'n_particles': 10, 'n_iterations': 10, 'compute_bands': [False]+[True]*(len(QSO_im_list)-1)}],
['PSO', {'sigma_scale': 0.8, 'n_particles': 100, 'n_iterations': 200, 'compute_bands': [True]*len(QSO_im_list)}],
['MCMC', {'n_burn': 10, 'n_run': 20, 'walkerRatio': 50, 'sigma_scale': .1}]
]
elif deep_seed == True:
fitting_kwargs_list = [
['PSO', {'sigma_scale': 0.8, 'n_particles': 150, 'n_iterations': 60, 'compute_bands': [True]+[False]*(len(QSO_im_list)-1)}],
['align_images', {'n_particles': 20, 'n_iterations': 20, 'compute_bands': [False]+[True]*(len(QSO_im_list)-1)}],
['PSO', {'sigma_scale': 0.8, 'n_particles': 150, 'n_iterations': 200, 'compute_bands': [True]*len(QSO_im_list)}],
['MCMC', {'n_burn': 20, 'n_run': 40, 'walkerRatio': 50, 'sigma_scale': .1}]
]
if no_MCMC == True:
del fitting_kwargs_list[-1]
start_time = time.time()
# lens_result, source_result, lens_light_result, ps_result, cosmo_temp, chain_list, param_list, samples_mcmc, param_mcmc, dist_mcmc = fitting_seq.fit_sequence(fitting_kwargs_list)
chain_list, param_list, samples_mcmc, param_mcmc, dist_mcmc = fitting_seq.fit_sequence(fitting_kwargs_list)
lens_result, source_result, lens_light_result, ps_result, cosmo_temp = fitting_seq.best_fit()
end_time = time.time()
print(end_time - start_time, 'total time needed for computation')
print('============ CONGRATULATION, YOUR JOB WAS SUCCESSFUL ================ ')
source_result_list, ps_result_list = [], []
image_reconstructed_list, error_map_list, image_ps_list, image_host_list, shift_RADEC_list=[], [], [], [],[]
imageLinearFit_list = []
for k in range(len(QSO_im_list)):
# this is the linear inversion. The kwargs will be updated afterwards
imageLinearFit_k = ImageLinearFit(data_class_list[k], psf_class_list[k], source_model_class=lightModel,
point_source_class=pointSource, kwargs_numerics=kwargs_numerics_list[k])
image_reconstructed_k, error_map_k, _, _ = imageLinearFit_k.image_linear_solve(kwargs_source=source_result, kwargs_ps=ps_result)
imageLinearFit_list.append(imageLinearFit_k)
[kwargs_data_k, kwargs_psf_k, kwargs_numerics_k] = fitting_seq.multi_band_list[k]
# data_class_k = data_class_list[k] #ImageData(**kwargs_data_k)
# psf_class_k = psf_class_list[k] #PSF(**kwargs_psf_k)
# imageModel_k = ImageModel(data_class_k, psf_class_k, source_model_class=lightModel,
# point_source_class=pointSource, kwargs_numerics=kwargs_numerics_list[k])
imageModel_k = imageModel_list[k]
modelPlot = ModelPlot(multi_band_list[k], kwargs_model, lens_result, source_result,
lens_light_result, ps_result, arrow_size=0.02, cmap_string="gist_heat", likelihood_mask=QSO_im_list[k])
print("source_result", 'for', "k", source_result)
image_host_k = []
for i in range(len(source_result)):
image_host_k.append(imageModel_list[k].source_surface_brightness(source_result,de_lensed=True,unconvolved=False, k=i))
image_ps_k = imageModel_k.point_source(ps_result)
# let's plot the output of the PSO minimizer
image_reconstructed_list.append(image_reconstructed_k)
source_result_list.append(source_result)
ps_result_list.append(ps_result)
error_map_list.append(error_map_k)
image_ps_list.append(image_ps_k)
image_host_list.append(image_host_k)
if 'ra_shift' in fitting_seq.multi_band_list[k][0].keys():
shift_RADEC_list.append([fitting_seq.multi_band_list[k][0]['ra_shift'], fitting_seq.multi_band_list[k][0]['dec_shift']])
else:
shift_RADEC_list.append([0,0])
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()
if tag is not None:
f.savefig('{0}_fitted_image_band{1}.pdf'.format(tag,new_band_seq[k]))
if pltshow == 0:
plt.close()
else:
plt.show()
if corner_plot==True and no_MCMC==False and k ==0:
# 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))
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=fixed_ps, **kwargs_constraints)
mcmc_new_list = []
labels_new = [r"Quasar flux", r"host_flux", r"source_x", r"source_y"]
# transform the parameter position of the MCMC chain in a lenstronomy convention with keyword arguments #
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.getParams(samples_mcmc[i+ len(samples_mcmc)/10*9])
image_reconstructed, _, _, _ = imageLinearFit_list[k].image_linear_solve(kwargs_source=kwargs_light_source_out, kwargs_ps=kwargs_ps_out)
image_ps = imageModel_list[k].point_source(kwargs_ps_out)
flux_quasar = np.sum(image_ps)
image_disk = imageModel_list[k].source_surface_brightness(kwargs_light_source_out,de_lensed=True,unconvolved=False, k=0)
flux_disk = np.sum(image_disk)
source_x = kwargs_ps_out[0]['ra_image']
source_y = kwargs_ps_out[0]['dec_image']
if flux_disk>0:
mcmc_new_list.append([flux_quasar, flux_disk, source_x, source_y])
plot = corner.corner(mcmc_new_list, labels=labels_new, show_titles=True)
if tag is not None:
plot.savefig('{0}_HOSTvsQSO_corner_band{1}.pdf'.format(tag,new_band_seq[k]))
if pltshow == 0:
plt.close()
else:
plt.show()
errp_list = []
for k in range(len(QSO_im_list)):
if QSO_std_list is None:
errp_list.append(np.sqrt(data_class_list[k].C_D+np.abs(error_map_list[k])))
else:
errp_list.append(np.sqrt(QSO_std_list[k]**2+np.abs(error_map_list[k])))
return source_result_list, ps_result_list, image_ps_list, image_host_list, errp_list, shift_RADEC_list, fitting_seq #fitting_seq.multi_band_list
