def test__log_likelihood_via_analysis__matches_manual_fit(
self, imaging_ci_7x7, pre_cti_data_7x7, traps_x1, ccd,
parallel_clocker_2d):
model = af.CollectionPriorModel(
cti=af.Model(ac.CTI2D, parallel_traps=traps_x1, parallel_ccd=ccd),
hyper_noise=af.Model(ac.ci.HyperCINoiseCollection),
)
analysis = ac.AnalysisImagingCI(dataset_ci=imaging_ci_7x7,
clocker=parallel_clocker_2d)
instance = model.instance_from_unit_vector([])
log_likelihood_via_analysis = analysis.log_likelihood_function(
instance=instance)
post_cti_data = parallel_clocker_2d.add_cti(
data=pre_cti_data_7x7.native,
parallel_trap_list=traps_x1,
parallel_ccd=ccd)
fit = ac.ci.FitImagingCI(dataset=analysis.dataset_ci,
post_cti_data=post_cti_data)
assert fit.log_likelihood == log_likelihood_via_analysis
def test__fits_to_extracted_and_full_datasets_available(
self,
imaging_ci_7x7,
mask_2d_7x7_unmasked,
parallel_clocker_2d,
samples_with_result,
):
masked_imaging_ci = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
masked_imaging_ci = masked_imaging_ci.apply_settings(
settings=ac.ci.SettingsImagingCI(parallel_pixels=(0, 1)))
analysis = ac.AnalysisImagingCI(dataset_ci=masked_imaging_ci,
clocker=parallel_clocker_2d)
result = ResultImagingCI(samples=samples_with_result,
analysis=analysis,
model=None,
search=None)
assert (result.max_log_likelihood_fit.mask == np.full(
fill_value=False, shape=(7, 1))).all()
assert (result.max_log_likelihood_full_fit.mask == np.full(
fill_value=False, shape=(7, 7))).all()
def test__noise_scaling_map_list_is_setup_correctly(
self,
imaging_ci_7x7,
mask_2d_7x7_unmasked,
layout_ci_7x7,
parallel_clocker_2d,
samples_with_result,
):
imaging_ci_7x7.cosmic_ray_map = None
masked_imaging_ci_7x7 = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
analysis = ac.AnalysisImagingCI(dataset_ci=masked_imaging_ci_7x7,
clocker=parallel_clocker_2d)
result = ResultImagingCI(samples=samples_with_result,
analysis=analysis,
model=None,
search=None)
assert (result.noise_scaling_map_list[0] ==
result.noise_scaling_map_of_regions_ci).all()
assert (result.noise_scaling_map_list[1] ==
result.noise_scaling_map_of_parallel_epers).all()
assert (result.noise_scaling_map_list[2] ==
result.noise_scaling_map_of_serial_trails).all()
assert (result.noise_scaling_map_list[3] ==
result.noise_scaling_map_of_serial_overscan_no_trails).all()
def test__parallel_and_serial_checks_raise_exception(self, imaging_ci_7x7):
model = af.CollectionPriorModel(cti=af.Model(
ac.CTI2D,
parallel_traps=[
ac.TrapInstantCapture(density=1.1),
ac.TrapInstantCapture(density=1.1),
],
parallel_ccd=ac.CCDPhase(),
))
analysis = ac.AnalysisImagingCI(
dataset_ci=imaging_ci_7x7,
clocker=None,
settings_cti=ac.SettingsCTI2D(parallel_total_density_range=(1.0,
2.0)),
)
instance = model.instance_from_prior_medians()
with pytest.raises(exc.PriorException):
analysis.log_likelihood_function(instance=instance)
model = af.CollectionPriorModel(cti=af.Model(
ac.CTI2D,
serial_traps=[
ac.TrapInstantCapture(density=1.1),
ac.TrapInstantCapture(density=1.1),
],
serial_ccd=ac.CCDPhase(),
))
analysis = ac.AnalysisImagingCI(
dataset_ci=[imaging_ci_7x7],
clocker=None,
settings_cti=ac.SettingsCTI2D(serial_total_density_range=(1.0,
2.0)),
)
instance = model.instance_from_prior_medians()
with pytest.raises(exc.PriorException):
analysis.log_likelihood_function(instance=instance)
def test__make_result__result_imaging_is_returned(self, imaging_ci_7x7,
pre_cti_data_7x7,
traps_x1, ccd,
parallel_clocker_2d):
model = af.CollectionPriorModel(
cti=af.Model(ac.CTI2D, parallel_traps=traps_x1, parallel_ccd=ccd),
hyper_noise=af.Model(ac.ci.HyperCINoiseCollection),
)
analysis = ac.AnalysisImagingCI(dataset_ci=imaging_ci_7x7,
clocker=parallel_clocker_2d)
search = mock.MockSearch(name="test_search")
result = search.fit(model=model, analysis=analysis)
assert isinstance(result, ResultImagingCI)
def test__full_and_extracted_fits_from_instance_and_imaging_ci(
self, imaging_ci_7x7, mask_2d_7x7_unmasked, traps_x1, ccd,
parallel_clocker_2d):
model = af.CollectionPriorModel(
cti=af.Model(ac.CTI2D, parallel_traps=traps_x1, parallel_ccd=ccd),
hyper_noise=af.Model(ac.ci.HyperCINoiseCollection),
)
masked_imaging_ci = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
masked_imaging_ci = masked_imaging_ci.apply_settings(
settings=ac.ci.SettingsImagingCI(parallel_pixels=(0, 1)))
post_cti_data = parallel_clocker_2d.add_cti(
data=masked_imaging_ci.pre_cti_data,
parallel_trap_list=traps_x1,
parallel_ccd=ccd,
)
analysis = ac.AnalysisImagingCI(dataset_ci=masked_imaging_ci,
clocker=parallel_clocker_2d)
instance = model.instance_from_unit_vector([])
fit_analysis = analysis.fit_from_instance(instance=instance)
fit = ac.ci.FitImagingCI(dataset=masked_imaging_ci,
post_cti_data=post_cti_data)
assert fit.image.shape == (7, 1)
assert fit_analysis.log_likelihood == pytest.approx(fit.log_likelihood)
fit_full_analysis = analysis.fit_full_dataset_from_instance(
instance=instance)
fit = ac.ci.FitImagingCI(dataset=imaging_ci_7x7,
post_cti_data=post_cti_data)
assert fit.image.shape == (7, 7)
assert fit_full_analysis.log_likelihood == pytest.approx(
fit.log_likelihood)
def test__extracted_fits_from_instance_and_imaging_ci__include_noise_scaling(
self,
imaging_ci_7x7,
mask_2d_7x7_unmasked,
traps_x1,
ccd,
parallel_clocker_2d,
layout_ci_7x7,
):
model = af.CollectionPriorModel(
cti=af.Model(ac.CTI2D, parallel_traps=traps_x1, parallel_ccd=ccd),
hyper_noise=af.Model(ac.ci.HyperCINoiseCollection,
regions_ci=ac.ci.HyperCINoiseScalar),
)
noise_scaling_map_list_list_of_regions_ci = [
ac.Array2D.ones(shape_native=(7, 7), pixel_scales=1.0)
]
imaging_ci_7x7.noise_scaling_map_list = [
noise_scaling_map_list_list_of_regions_ci[0]
]
masked_imaging_ci = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
masked_imaging_ci = masked_imaging_ci.apply_settings(
settings=ac.ci.SettingsImagingCI(parallel_pixels=(0, 1)))
analysis = ac.AnalysisImagingCI(dataset_ci=masked_imaging_ci,
clocker=parallel_clocker_2d)
instance = model.instance_from_prior_medians()
fit_analysis = analysis.fit_from_instance(instance=instance,
hyper_noise_scale=True)
post_cti_data = parallel_clocker_2d.add_cti(
data=masked_imaging_ci.pre_cti_data,
parallel_trap_list=traps_x1,
parallel_ccd=ccd,
)
fit = ac.ci.FitImagingCI(
dataset=masked_imaging_ci,
post_cti_data=post_cti_data,
hyper_noise_scalars=[instance.hyper_noise.regions_ci],
)
assert fit.image.shape == (7, 1)
assert fit.log_likelihood == pytest.approx(fit_analysis.log_likelihood,
1.0e-4)
fit = ac.ci.FitImagingCI(
dataset=masked_imaging_ci,
post_cti_data=post_cti_data,
hyper_noise_scalars=[ac.ci.HyperCINoiseScalar(scale_factor=0.0)],
)
assert fit.log_likelihood != pytest.approx(fit_analysis.log_likelihood,
1.0e-4)
fit_full_analysis = analysis.fit_full_dataset_from_instance(
instance=instance, hyper_noise_scale=True)
masked_imaging_ci = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
fit = ac.ci.FitImagingCI(
dataset=masked_imaging_ci,
post_cti_data=post_cti_data,
hyper_noise_scalars=[instance.hyper_noise.regions_ci],
)
assert fit.image.shape == (7, 7)
assert fit.log_likelihood == pytest.approx(
fit_full_analysis.log_likelihood, 1.0e-4)
fit = ac.ci.FitImagingCI(
dataset=masked_imaging_ci,
post_cti_data=post_cti_data,
hyper_noise_scalars=[ac.ci.HyperCINoiseScalar(scale_factor=0.0)],
)
assert fit.log_likelihood != pytest.approx(
fit_full_analysis.log_likelihood, 1.0e-4)
imaging_ci_trimmed_list = [
imaging_ci.apply_mask(mask=mask_ci) for imaging_ci in imaging_ci_list
]
imaging_ci_trimmed_list = [
imaging_ci.apply_settings(settings=ac.ci.SettingsImagingCI(parallel_columns=(0, 1)))
for imaging_ci in imaging_ci_trimmed_list
]
"""
__Analysis__
The `AnalysisImagingCI` object defines the `log_likelihood_function` used by the non-linear search to fit the model to
the `ImagingCI`dataset.
"""
analysis = ac.AnalysisImagingCI(dataset_ci=imaging_ci_trimmed_list, clocker=clocker)
analysis_list = [
ac.AnalysisImagingCI(dataset_ci=imaging_ci, clocker=clocker)
for imaging_ci in imaging_ci_trimmed_list
]
analysis = analysis_list[0] + analysis_list[1] + analysis_list[2] + analysis_list[3]
"""
__Model-Fit__
We can now begin the model-fit by passing the model and analysis object to the search, which performs a non-linear
search to find which models fit the data with the highest likelihood.
Checkout the folder `autocti_workspace/output/imaging_ci/parallel[x2]` for live outputs
of the results of the fit, including on-the-fly visualization of the best fit model!
def test__noise_scaling_map_list_is_list_of_result__are_correct(
self,
imaging_ci_7x7,
mask_2d_7x7_unmasked,
parallel_clocker_2d,
layout_ci_7x7,
samples_with_result,
traps_x1,
ccd,
):
noise_scaling_map_list_list_of_regions_ci = [
ac.Array2D.ones(shape_native=(7, 7), pixel_scales=1.0)
]
noise_scaling_map_list_list_of_parallel_epers = [
ac.Array2D.full(fill_value=2.0,
shape_native=(7, 7),
pixel_scales=1.0)
]
noise_scaling_map_list_list_of_serial_trails = [
ac.Array2D.full(fill_value=3.0,
shape_native=(7, 7),
pixel_scales=1.0)
]
noise_scaling_map_list_list_of_serial_overscan_no_trails = [
ac.Array2D.full(fill_value=4.0,
shape_native=(7, 7),
pixel_scales=1.0)
]
imaging_ci_7x7.noise_scaling_map_list = [
noise_scaling_map_list_list_of_regions_ci[0],
noise_scaling_map_list_list_of_parallel_epers[0],
noise_scaling_map_list_list_of_serial_trails[0],
noise_scaling_map_list_list_of_serial_overscan_no_trails[0],
]
masked_imaging_ci_7x7 = imaging_ci_7x7.apply_mask(
mask=mask_2d_7x7_unmasked)
analysis = ac.AnalysisImagingCI(dataset_ci=masked_imaging_ci_7x7,
clocker=parallel_clocker_2d)
fit_analysis = analysis.fit_from_instance(
instance=samples_with_result.max_log_likelihood_instance)
result = ResultImagingCI(samples=samples_with_result,
analysis=analysis,
model=None,
search=None)
assert result.noise_scaling_map_of_regions_ci == pytest.approx(
fit_analysis.chi_squared_map_of_regions_ci, 1.0e-2)
assert result.noise_scaling_map_of_parallel_epers == pytest.approx(
fit_analysis.chi_squared_map_of_parallel_epers, 1.0e-2)
assert result.noise_scaling_map_of_serial_trails == pytest.approx(
fit_analysis.chi_squared_map_of_serial_trails, 1.0e-2)
assert result.noise_scaling_map_of_serial_overscan_no_trails == pytest.approx(
fit_analysis.chi_squared_map_of_serial_overscan_no_trails, 1.0e-2)
assert result.noise_scaling_map_of_regions_ci[1, 1] == pytest.approx(
18.16, 1.0e-1)
assert result.noise_scaling_map_of_parallel_epers[1,
1] == pytest.approx(
0.0, 1.0e-4)
assert result.noise_scaling_map_of_serial_trails[1,
1] == pytest.approx(
0.0, 1.0e-4)
assert result.noise_scaling_map_of_serial_overscan_no_trails[
1, 1] == pytest.approx(0.0, 1.0e-4)
model = af.CollectionPriorModel(
cti=af.Model(
ac.CTI2D,
parallel_traps=traps_x1,
parallel_ccd=ccd,
serial_traps=traps_x1,
serial_ccd=ccd,
),
hyper_noise=af.Model(
ac.ci.HyperCINoiseCollection,
regions_ci=ac.ci.HyperCINoiseScalar(scale_factor=1.0),
parallel_epers=ac.ci.HyperCINoiseScalar(scale_factor=1.0),
serial_trails=ac.ci.HyperCINoiseScalar(scale_factor=1.0),
serial_overscan_no_trails=ac.ci.HyperCINoiseScalar(
scale_factor=1.0),
),
)
instance = model.instance_from_prior_medians()
fit_analysis = analysis.fit_from_instance(instance=instance)
assert result.noise_scaling_map_of_regions_ci != pytest.approx(
fit_analysis.chi_squared_map_of_regions_ci, 1.0e-2)
assert result.noise_scaling_map_of_parallel_epers != pytest.approx(
fit_analysis.chi_squared_map_of_parallel_epers, 1.0e-2)
assert result.noise_scaling_map_of_serial_trails != pytest.approx(
fit_analysis.chi_squared_map_of_serial_trails, 1.0e-2)
assert result.noise_scaling_map_of_serial_overscan_no_trails != pytest.approx(
fit_analysis.chi_squared_map_of_serial_overscan_no_trails, 1.0e-2)
"""
__Search + Analysis + Model-Fit (Search 1)__
To reduce run-times, we trim the `ImagingCI` data from the high resolution data (e.g. 2000 columns) to just 50 columns
to speed up the model-fit at the expense of inferring larger errors on the CTI model.
"""
search = af.DynestyStatic(path_prefix=path_prefix,
name="search[1]_parallel[x1]",
nlive=50)
imaging_ci_trimmed_list = [
imaging_ci.apply_settings(settings=ac.ci.SettingsImagingCI(
parallel_columns=(0, 50))) for imaging_ci in imaging_ci_list
]
analysis = ac.AnalysisImagingCI(dataset_ci=imaging_ci_trimmed_list,
clocker=parallel_clocker)
result_1 = search.fit(model=model, analysis=analysis)
"""
__Model (Search 2)__
We use the results of search 1 to create the lens model fitted in search 2, with:
- Two or more parallel `TrapInstantCapture`'s species [4+ parameters: prior on density initialized from search 1].
- A simple `CCD` volume filling parametrization with fixed notch depth and capacity [1 parameter: priors initialized
from search 1].
The number of free parameters and therefore the dimensionality of non-linear parameter space is N=5 or more.
"""
parallel_trap_0 = af.Model(ac.TrapInstantCapture)