def setup(self):
self.nbins = 30
binning = np.logspace(-1, 1, self.nbins + 1) * u.TeV
self.source_model = PowerLawSpectralModel(index=2.1,
amplitude=1e5 *
u.Unit("cm-2 s-1 TeV-1"),
reference=0.1 * u.TeV)
self.livetime = 100 * u.s
aeff = EffectiveAreaTable.from_constant(binning, "1 cm2")
bkg_rate = np.ones(self.nbins) / u.s
bkg_expected = (bkg_rate * self.livetime).to_value("")
self.bkg = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=bkg_expected)
random_state = get_random_state(23)
flux = self.source_model.integral(binning[:-1], binning[1:])
self.npred = (flux * aeff.data.data[0] * self.livetime).to_value("")
self.npred += bkg_expected
source_counts = random_state.poisson(self.npred)
self.src = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=source_counts)
self.dataset = SpectrumDataset(
model=self.source_model,
counts=self.src,
aeff=aeff,
livetime=self.livetime,
background=self.bkg,
)
def test_spectrum_dataset_stack_nondiagonal_no_bkg(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp1 = EDispKernel.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.1, 0.0)
livetime = self.livetime
dataset1 = SpectrumDataset(counts=None,
livetime=livetime,
aeff=aeff,
edisp=edisp1,
background=None)
livetime2 = livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:], aeff.data.data)
edisp2 = EDispKernel.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.2, 0.0)
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp2,
background=None,
)
dataset1.stack(dataset2)
assert dataset1.counts is None
assert dataset1.background is None
assert dataset1.livetime == 2 * self.livetime
assert_allclose(dataset1.aeff.data.data.to_value("m2"),
aeff.data.data.to_value("m2"))
assert_allclose(dataset1.edisp.get_bias(1 * u.TeV), 0.0, atol=1.2e-3)
assert_allclose(dataset1.edisp.get_resolution(1 * u.TeV),
0.1581,
atol=1e-2)
def setup(self):
self.nbins = 30
binning = np.logspace(-1, 1, self.nbins + 1) * u.TeV
self.source_model = PowerLawSpectralModel(index=2.1,
amplitude=1e5 / u.TeV / u.s,
reference=0.1 * u.TeV)
self.livetime = 100 * u.s
bkg_rate = np.ones(self.nbins) / u.s
bkg_expected = bkg_rate * self.livetime
self.bkg = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=bkg_expected)
random_state = get_random_state(23)
self.npred = (self.source_model.integral(binning[:-1], binning[1:]) *
self.livetime)
self.npred += bkg_expected
source_counts = random_state.poisson(self.npred)
self.src = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=source_counts)
self.dataset = SpectrumDataset(
model=self.source_model,
counts=self.src,
livetime=self.livetime,
background=self.bkg,
)
def test_set_model(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EnergyDispersion.from_diagonal_response(
self.src.energy.edges, self.src.energy.edges)
dataset = SpectrumDataset(None, self.src, self.livetime, None, aeff,
edisp, self.bkg)
with pytest.raises(AttributeError):
dataset.parameters
dataset.model = self.source_model
assert dataset.parameters[0] == self.source_model.parameters[0]
def spectrum_dataset_crab_fine():
e_true = np.logspace(-2, 2.5, 109) * u.TeV
e_reco = np.logspace(-2, 2, 73) * u.TeV
pos = SkyCoord(83.63, 22.01, unit="deg", frame="icrs")
radius = Angle(0.11, "deg")
region = CircleSkyRegion(pos, radius)
return SpectrumDataset.create(e_reco, e_true, region=region)
def test_set_model(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EDispKernel.from_diagonal_response(self.src.energy.edges,
self.src.energy.edges)
dataset = SpectrumDataset(None, self.src, self.livetime, None, aeff,
edisp, self.bkg)
spectral_model = PowerLawSpectralModel()
model = SkyModel(spectral_model=spectral_model, name="test")
dataset.models = model
assert dataset.models["test"] is model
models = Models([model])
dataset.models = models
assert dataset.models["test"] is model
def spectrum_dataset_gc():
e_reco = np.logspace(0, 2, 5) * u.TeV
e_true = np.logspace(-0.5, 2, 11) * u.TeV
pos = SkyCoord(0.0, 0.0, unit="deg", frame="galactic")
radius = Angle(0.11, "deg")
region = CircleSkyRegion(pos, radius)
return SpectrumDataset.create(e_reco, e_true, region=region)
def test_spectrum_dataset_stack_diagonal_safe_mask(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EDispKernel.from_diagonal_response(self.src.energy.edges,
self.src.energy.edges)
livetime = self.livetime
dataset1 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime,
aeff=aeff,
edisp=edisp,
background=self.bkg.copy(),
)
livetime2 = 0.5 * livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:],
2 * aeff.data.data)
bkg2 = CountsSpectrum(
self.src.energy.edges[:-1],
self.src.energy.edges[1:],
data=2 * self.bkg.data,
)
safe_mask2 = np.ones_like(self.src.data, bool)
safe_mask2[0] = False
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp,
background=bkg2,
mask_safe=safe_mask2,
)
dataset1.stack(dataset2)
assert_allclose(dataset1.counts.data[1:], self.src.data[1:] * 2)
assert_allclose(dataset1.counts.data[0], self.src.data[0])
assert dataset1.livetime == 1.5 * self.livetime
assert_allclose(dataset1.background.data[1:], 3 * self.bkg.data[1:])
assert_allclose(dataset1.background.data[0], self.bkg.data[0])
assert_allclose(
dataset1.aeff.data.data.to_value("m2"),
4.0 / 3 * aeff.data.data.to_value("m2"),
)
assert_allclose(dataset1.edisp.pdf_matrix[1:], edisp.pdf_matrix[1:])
assert_allclose(dataset1.edisp.pdf_matrix[0],
0.5 * edisp.pdf_matrix[0])
def test_stack_no_livetime():
e_reco = np.logspace(0, 1, 3) * u.TeV
dataset_1 = SpectrumDataset.create(e_reco=e_reco)
dataset_1.livetime = None
dataset_2 = dataset_1.copy()
with pytest.raises(ValueError):
dataset_1.stack(dataset_2)
def _spectrum_extraction(self):
"""Run all steps for the spectrum extraction."""
log.info("Reducing spectrum datasets.")
datasets_settings = self.config.datasets
on_lon = datasets_settings.on_region.lon
on_lat = datasets_settings.on_region.lat
on_center = SkyCoord(on_lon,
on_lat,
frame=datasets_settings.on_region.frame)
on_region = CircleSkyRegion(on_center,
datasets_settings.on_region.radius)
maker_config = {}
if datasets_settings.containment_correction:
maker_config[
"containment_correction"] = datasets_settings.containment_correction
e_reco = self._make_energy_axis(
datasets_settings.geom.axes.energy).edges
maker_config["selection"] = ["counts", "aeff", "edisp"]
dataset_maker = SpectrumDatasetMaker(**maker_config)
bkg_maker_config = {}
if datasets_settings.background.exclusion:
exclusion_region = Map.read(datasets_settings.background.exclusion)
bkg_maker_config["exclusion_mask"] = exclusion_region
bkg_maker = ReflectedRegionsBackgroundMaker(**bkg_maker_config)
safe_mask_selection = self.config.datasets.safe_mask.methods
safe_mask_settings = self.config.datasets.safe_mask.settings
safe_mask_maker = SafeMaskMaker(methods=safe_mask_selection,
**safe_mask_settings)
e_true = self._make_energy_axis(
datasets_settings.geom.axes.energy_true).edges
reference = SpectrumDataset.create(e_reco=e_reco,
e_true=e_true,
region=on_region)
datasets = []
for obs in self.observations:
log.info(f"Processing observation {obs.obs_id}")
dataset = dataset_maker.run(reference.copy(), obs)
dataset = bkg_maker.run(dataset, obs)
if dataset.counts_off is None:
log.info(
f"No OFF region found for observation {obs.obs_id}. Discarding."
)
continue
dataset = safe_mask_maker.run(dataset, obs)
log.debug(dataset)
datasets.append(dataset)
self.datasets = Datasets(datasets)
if self.config.datasets.stack:
stacked = self.datasets.stack_reduce(name="stacked")
self.datasets = Datasets([stacked])
def test_cash(self):
"""Simple CASH fit to the on vector"""
dataset = SpectrumDataset(model=self.source_model, counts=self.src)
npred = dataset.npred().data
assert_allclose(npred[5], 660.5171, rtol=1e-5)
stat_val = dataset.likelihood()
assert_allclose(stat_val, -107346.5291, rtol=1e-5)
self.source_model.parameters["index"].value = 1.12
fit = Fit([dataset])
result = fit.run()
# These values are check with sherpa fits, do not change
pars = result.parameters
assert_allclose(pars["index"].value, 1.995525, rtol=1e-3)
assert_allclose(pars["amplitude"].value, 100245.9, rtol=1e-3)
def test_incorrect_mask(self):
mask_fit = np.ones(self.nbins, dtype=np.dtype("float"))
with pytest.raises(ValueError):
SpectrumDataset(
models=self.source_model,
counts=self.src,
livetime=self.livetime,
mask_fit=mask_fit,
background=self.bkg,
)
def data(self) -> SpectrumDataset:
"""Actual event data in form of a SpectrumDataset.
"""
dataset_empty = SpectrumDataset.create(e_reco=self.energy_axis,
e_true=self.energy_axis,
region=self.on_region)
maker = SpectrumDatasetMaker(containment_correction=False,
selection=["background", "aeff", "edisp"])
dataset = maker.run(dataset_empty, self.obs)
dataset.models = self.true_model
dataset.fake()
return dataset
def test_likelihood_profile(self):
dataset = SpectrumDataset(
model=self.source_model,
counts=self.src,
mask_safe=np.ones(self.src.energy.nbin, dtype=bool),
)
fit = Fit([dataset])
result = fit.run()
true_idx = result.parameters["index"].value
values = np.linspace(0.95 * true_idx, 1.05 * true_idx, 100)
profile = fit.likelihood_profile("index", values=values)
actual = values[np.argmin(profile["likelihood"])]
assert_allclose(actual, true_idx, rtol=0.01)
def spectrum_dataset():
e_true = np.logspace(0, 1, 21) * u.TeV
e_reco = np.logspace(0, 1, 5) * u.TeV
aeff = EffectiveAreaTable.from_constant(value=1e6 * u.m**2, energy=e_true)
edisp = EDispKernel.from_diagonal_response(e_true, e_reco)
data = 3600 * np.ones(4)
data[-1] *= 1e-3
background = CountsSpectrum(energy_lo=e_reco[:-1],
energy_hi=e_reco[1:],
data=data)
return SpectrumDataset(aeff=aeff,
livetime="1h",
edisp=edisp,
background=background)
def test_spectrumdataset_create(self):
e_reco = u.Quantity([0.1, 1, 10.0], "TeV")
e_true = u.Quantity([0.05, 0.5, 5, 20.0], "TeV")
empty_dataset = SpectrumDataset.create(e_reco, e_true)
assert empty_dataset.counts.total_counts == 0
assert empty_dataset.data_shape[0] == 2
assert empty_dataset.background.total_counts == 0
assert empty_dataset.background.energy.nbin == 2
assert empty_dataset.aeff.data.axis("energy").nbin == 3
assert empty_dataset.edisp.data.axis("e_reco").nbin == 2
assert empty_dataset.livetime.value == 0
assert len(empty_dataset.gti.table) == 0
assert empty_dataset.energy_range[0] is None
assert_allclose(empty_dataset.mask_safe, 0)
def test_npred_models(self):
e_reco = MapAxis.from_energy_bounds("1 TeV", "10 TeV", nbin=3).edges
dataset = SpectrumDataset.create(e_reco=e_reco)
dataset.livetime = 1 * u.h
dataset.aeff.data.data += 1e10 * u.Unit("cm2")
pwl_1 = PowerLawSpectralModel(index=2)
pwl_2 = PowerLawSpectralModel(index=2)
model_1 = SkyModel(spectral_model=pwl_1)
model_2 = SkyModel(spectral_model=pwl_2)
dataset.models = Models([model_1, model_2])
npred = dataset.npred()
assert_allclose(npred.data.sum(), 64.8)
def test_run(observations, phase_bkg_maker):
maker = SpectrumDatasetMaker()
e_reco = np.logspace(0, 2, 5) * u.TeV
e_true = np.logspace(-0.5, 2, 11) * u.TeV
pos = SkyCoord("08h35m20.65525s", "-45d10m35.1545s", frame="icrs")
radius = Angle(0.2, "deg")
region = SphericalCircleSkyRegion(pos, radius)
dataset_empty = SpectrumDataset.create(e_reco, e_true, region=region)
obs = observations["111630"]
dataset = maker.run(dataset_empty, obs)
dataset_on_off = phase_bkg_maker.run(dataset, obs)
assert_allclose(dataset_on_off.acceptance, 0.1)
assert_allclose(dataset_on_off.acceptance_off, 0.3)
assert_allclose(dataset_on_off.counts.data.sum(), 28)
assert_allclose(dataset_on_off.counts_off.data.sum(), 57)
def test_reflected_bkg_maker_no_off(reflected_bkg_maker, observations):
pos = SkyCoord(83.6333313, 21.51444435, unit="deg", frame="icrs")
radius = Angle(0.11, "deg")
region = CircleSkyRegion(pos, radius)
maker = SpectrumDatasetMaker(selection=["counts"])
datasets = []
e_reco = np.logspace(0, 2, 5) * u.TeV
e_true = np.logspace(-0.5, 2, 11) * u.TeV
dataset_empty = SpectrumDataset.create(e_reco=e_reco,
e_true=e_true,
region=region)
for obs in observations:
dataset = maker.run(dataset_empty, obs)
dataset_on_off = reflected_bkg_maker.run(dataset, obs)
datasets.append(dataset_on_off)
assert datasets[0].counts_off is None
assert_allclose(datasets[0].acceptance_off, 0)
def test_reflected_bkg_maker(on_region, reflected_bkg_maker, observations):
datasets = []
e_reco = np.logspace(0, 2, 5) * u.TeV
e_true = np.logspace(-0.5, 2, 11) * u.TeV
dataset_empty = SpectrumDataset.create(e_reco=e_reco,
e_true=e_true,
region=on_region)
maker = SpectrumDatasetMaker(selection=["counts"])
for obs in observations:
dataset = maker.run(dataset_empty, obs)
dataset_on_off = reflected_bkg_maker.run(dataset, obs)
datasets.append(dataset_on_off)
assert_allclose(datasets[0].counts_off.data.sum(), 76)
assert_allclose(datasets[1].counts_off.data.sum(), 60)
regions_0 = compound_region_to_list(datasets[0].counts_off.region)
regions_1 = compound_region_to_list(datasets[1].counts_off.region)
assert_allclose(len(regions_0), 11)
assert_allclose(len(regions_1), 11)
def to_spectrum_dataset(self, on_region, containment_correction=False):
"""Return a ~gammapy.spectrum.SpectrumDataset from on_region.
Counts and background are summed in the on_region.
Effective area is taken from the average exposure divided by the livetime.
Here we assume it is the sum of the GTIs.
EnergyDispersion is obtained at the on_region center.
Only regions with centers are supported.
Parameters
----------
on_region : `~regions.SkyRegion`
the input ON region on which to extract the spectrum
containment_correction : bool
Apply containment correction for point sources and circular on regions
Returns
-------
dataset : `~gammapy.spectrum.SpectrumDataset`
the resulting reduced dataset
"""
if self.gti is not None:
livetime = self.gti.time_sum
else:
raise ValueError("No GTI in `MapDataset`, cannot compute livetime")
if self.counts is not None:
counts = self.counts.get_spectrum(on_region, np.sum)
else:
counts = None
if self.background_model is not None:
background = self.background_model.evaluate().get_spectrum(
on_region, np.sum
)
else:
background = None
if self.exposure is not None:
exposure = self.exposure.get_spectrum(on_region, np.mean)
aeff = EffectiveAreaTable(
energy_lo=exposure.energy.edges[:-1],
energy_hi=exposure.energy.edges[1:],
data=exposure.data / livetime,
)
else:
aeff = None
if containment_correction:
if not isinstance(on_region, CircleSkyRegion):
raise TypeError(
"Containement correction is only supported for"
" `CircleSkyRegion`."
)
elif self.psf is None or isinstance(self.psf, PSFKernel):
raise ValueError("No PSFMap set. Containement correction impossible")
else:
psf_table = self.psf.get_energy_dependent_table_psf(on_region.center)
aeff = apply_containment_fraction(aeff, psf_table, on_region.radius)
if self.edisp is not None:
if isinstance(self.edisp, EnergyDispersion):
edisp = self.edisp
else:
self.edisp.get_energy_dispersion(on_region.center, self._energy_axis)
else:
edisp = None
return SpectrumDataset(
counts=counts,
background=background,
aeff=aeff,
edisp=edisp,
livetime=livetime,
gti=self.gti,
name=self.name,
)
class TestSpectrumDataset:
"""Test fit on counts spectra without any IRFs"""
def setup(self):
self.nbins = 30
binning = np.logspace(-1, 1, self.nbins + 1) * u.TeV
self.source_model = PowerLaw(index=2.1,
amplitude=1e5 / u.TeV / u.s,
reference=0.1 * u.TeV)
self.livetime = 100 * u.s
bkg_rate = np.ones(self.nbins) / u.s
bkg_expected = bkg_rate * self.livetime
self.bkg = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=bkg_expected)
random_state = get_random_state(23)
self.npred = (self.source_model.integral(binning[:-1], binning[1:]) *
self.livetime)
self.npred += bkg_expected
source_counts = random_state.poisson(self.npred)
self.src = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=source_counts)
self.dataset = SpectrumDataset(
model=self.source_model,
counts=self.src,
livetime=self.livetime,
background=self.bkg,
)
def test_data_shape(self):
assert self.dataset.data_shape[0] == self.nbins
def test_energy_range(self):
energy_range = self.dataset.energy_range
assert energy_range.unit == u.TeV
assert_allclose(energy_range.to_value("TeV"), [0.1, 10.0])
def test_cash(self):
"""Simple CASH fit to the on vector"""
fit = Fit(self.dataset)
result = fit.run()
assert result.success
assert "minuit" in repr(result)
npred = self.dataset.npred().data.sum()
assert_allclose(npred, self.npred.sum(), rtol=1e-3)
assert_allclose(result.total_stat, -18087404.624, rtol=1e-3)
pars = result.parameters
assert_allclose(pars["index"].value, 2.1, rtol=1e-2)
assert_allclose(pars.error("index"), 0.00127, rtol=1e-2)
assert_allclose(pars["amplitude"].value, 1e5, rtol=1e-3)
assert_allclose(pars.error("amplitude"), 153.450, rtol=1e-2)
def test_fake(self):
"""Test the fake dataset"""
real_dataset = self.dataset.copy()
self.dataset.fake(314)
assert real_dataset.counts.data.shape == self.dataset.counts.data.shape
assert real_dataset.background.data.sum(
) == self.dataset.background.data.sum()
assert int(real_dataset.counts.data.sum()) == 907010
assert self.dataset.counts.data.sum() == 907331
def test_incorrect_mask(self):
mask_fit = np.ones(self.nbins, dtype=np.dtype("float"))
with pytest.raises(ValueError):
SpectrumDataset(
model=self.source_model,
counts=self.src,
livetime=self.livetime,
mask_fit=mask_fit,
background=self.bkg,
)
def test_set_model(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EnergyDispersion.from_diagonal_response(
self.src.energy.edges, self.src.energy.edges)
dataset = SpectrumDataset(None, self.src, self.livetime, None, aeff,
edisp, self.bkg)
with pytest.raises(AttributeError):
dataset.parameters
dataset.model = self.source_model
assert dataset.parameters[0] == self.source_model.parameters[0]
def test_str(self):
assert "SpectrumDataset" in str(self.dataset)
# ## Spatial analysis
#
# See other notebooks for how to run a 3D cube or 2D image based analysis.
# ## Spectrum
#
# We'll run a spectral analysis using the classical reflected regions background estimation method,
# and using the on-off (often called WSTAT) likelihood function.
# In[ ]:
e_reco = np.logspace(-1, np.log10(40), 40) * u.TeV
e_true = np.logspace(np.log10(0.05), 2, 200) * u.TeV
dataset_empty = SpectrumDataset.create(e_reco=e_reco,
e_true=e_true,
region=on_region)
# In[ ]:
dataset_maker = SpectrumDatasetMaker(containment_correction=False,
selection=["counts", "aeff", "edisp"])
bkg_maker = ReflectedRegionsBackgroundMaker(exclusion_mask=exclusion_mask)
safe_mask_masker = SafeMaskMaker(methods=["aeff-max"], aeff_percent=10)
# In[ ]:
get_ipython().run_cell_magic(
'time', '',
'datasets = []\n\nfor observation in observations:\n dataset = dataset_maker.run(dataset_empty, observation)\n dataset_on_off = bkg_maker.run(dataset, observation)\n dataset_on_off = safe_mask_masker.run(dataset_on_off, observation)\n datasets.append(dataset_on_off)'
)
class TestSpectrumDataset:
"""Test fit on counts spectra without any IRFs"""
def setup(self):
self.nbins = 30
binning = np.logspace(-1, 1, self.nbins + 1) * u.TeV
self.source_model = PowerLawSpectralModel(index=2.1,
amplitude=1e5 / u.TeV / u.s,
reference=0.1 * u.TeV)
self.livetime = 100 * u.s
bkg_rate = np.ones(self.nbins) / u.s
bkg_expected = bkg_rate * self.livetime
self.bkg = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=bkg_expected)
random_state = get_random_state(23)
self.npred = (self.source_model.integral(binning[:-1], binning[1:]) *
self.livetime)
self.npred += bkg_expected
source_counts = random_state.poisson(self.npred)
self.src = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=source_counts)
self.dataset = SpectrumDataset(
model=self.source_model,
counts=self.src,
livetime=self.livetime,
background=self.bkg,
)
def test_data_shape(self):
assert self.dataset.data_shape[0] == self.nbins
def test_energy_range(self):
energy_range = self.dataset.energy_range
assert energy_range.unit == u.TeV
assert_allclose(energy_range.to_value("TeV"), [0.1, 10.0])
def test_cash(self):
"""Simple CASH fit to the on vector"""
fit = Fit(self.dataset)
result = fit.run()
assert result.success
assert "minuit" in repr(result)
npred = self.dataset.npred().data.sum()
assert_allclose(npred, self.npred.sum(), rtol=1e-3)
assert_allclose(result.total_stat, -18087404.624, rtol=1e-3)
pars = result.parameters
assert_allclose(pars["index"].value, 2.1, rtol=1e-2)
assert_allclose(pars.error("index"), 0.00127, rtol=1e-2)
assert_allclose(pars["amplitude"].value, 1e5, rtol=1e-3)
assert_allclose(pars.error("amplitude"), 153.450, rtol=1e-2)
def test_fake(self):
"""Test the fake dataset"""
real_dataset = self.dataset.copy()
self.dataset.fake(314)
assert real_dataset.counts.data.shape == self.dataset.counts.data.shape
assert real_dataset.background.data.sum(
) == self.dataset.background.data.sum()
assert int(real_dataset.counts.data.sum()) == 907010
assert self.dataset.counts.data.sum() == 907331
def test_incorrect_mask(self):
mask_fit = np.ones(self.nbins, dtype=np.dtype("float"))
with pytest.raises(ValueError):
SpectrumDataset(
model=self.source_model,
counts=self.src,
livetime=self.livetime,
mask_fit=mask_fit,
background=self.bkg,
)
def test_set_model(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EnergyDispersion.from_diagonal_response(
self.src.energy.edges, self.src.energy.edges)
dataset = SpectrumDataset(None, self.src, self.livetime, None, aeff,
edisp, self.bkg)
with pytest.raises(AttributeError):
dataset.parameters
dataset.model = self.source_model
assert dataset.parameters[0] == self.source_model.parameters[0]
def test_str(self):
assert "SpectrumDataset" in str(self.dataset)
def test_spectrumdataset_create(self):
e_reco = u.Quantity([0.1, 1, 10.0], "TeV")
e_true = u.Quantity([0.05, 0.5, 5, 20.0], "TeV")
empty_dataset = SpectrumDataset.create(e_reco, e_true)
assert empty_dataset.counts.total_counts == 0
assert empty_dataset.data_shape[0] == 2
assert empty_dataset.background.total_counts == 0
assert empty_dataset.background.energy.nbin == 2
assert empty_dataset.aeff.data.axis("energy").nbin == 3
assert empty_dataset.edisp.data.axis("e_reco").nbin == 2
assert empty_dataset.livetime.value == 0
assert len(empty_dataset.gti.table) == 0
assert empty_dataset.energy_range[0] is None
assert_allclose(empty_dataset.mask_safe, 0)
def test_spectrum_dataset_stack_diagonal_safe_mask(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EnergyDispersion.from_diagonal_response(
self.src.energy.edges, self.src.energy.edges)
livetime = self.livetime
dataset1 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime,
aeff=aeff,
edisp=edisp,
background=self.bkg.copy(),
)
livetime2 = 0.5 * livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:],
2 * aeff.data.data)
bkg2 = CountsSpectrum(
self.src.energy.edges[:-1],
self.src.energy.edges[1:],
data=2 * self.bkg.data,
)
safe_mask2 = np.ones_like(self.src.data, bool)
safe_mask2[0] = False
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp,
background=bkg2,
mask_safe=safe_mask2,
)
dataset1.stack(dataset2)
assert_allclose(dataset1.counts.data[1:], self.src.data[1:] * 2)
assert_allclose(dataset1.counts.data[0], self.src.data[0])
assert dataset1.livetime == 1.5 * self.livetime
assert_allclose(dataset1.background.data[1:], 3 * self.bkg.data[1:])
assert_allclose(dataset1.background.data[0], self.bkg.data[0])
assert_allclose(
dataset1.aeff.data.data.to_value("m2"),
4.0 / 3 * aeff.data.data.to_value("m2"),
)
assert_allclose(dataset1.edisp.pdf_matrix[1:], edisp.pdf_matrix[1:])
assert_allclose(dataset1.edisp.pdf_matrix[0],
0.5 * edisp.pdf_matrix[0])
def test_spectrum_dataset_stack_nondiagonal_no_bkg(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp1 = EnergyDispersion.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.1, 0.0)
livetime = self.livetime
dataset1 = SpectrumDataset(counts=None,
livetime=livetime,
aeff=aeff,
edisp=edisp1,
background=None)
livetime2 = livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:], aeff.data.data)
edisp2 = EnergyDispersion.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.2, 0.0)
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp2,
background=None,
)
dataset1.stack(dataset2)
assert dataset1.counts is None
assert dataset1.background is None
assert dataset1.livetime == 2 * self.livetime
assert_allclose(dataset1.aeff.data.data.to_value("m2"),
aeff.data.data.to_value("m2"))
assert_allclose(dataset1.edisp.get_bias(1 * u.TeV), 0.0, atol=1e-3)
assert_allclose(dataset1.edisp.get_resolution(1 * u.TeV),
0.1581,
atol=1e-2)
mask = data_store.obs_table["TARGET_NAME"] == "Crab"
obs_ids = data_store.obs_table["OBS_ID"][mask].data
observations = data_store.get_observations(obs_ids)
crab_position = SkyCoord(83.63, 22.01, unit="deg", frame="icrs")
# The ON region center is defined in the icrs frame. The angle is defined w.r.t. to its axis.
rectangle = RectangleSkyRegion(center=crab_position,
width=0.5 * u.deg,
height=0.4 * u.deg,
angle=0 * u.deg)
bkg_maker = ReflectedRegionsBackgroundMaker(min_distance=0.1 * u.rad)
dataset_maker = SpectrumDatasetMaker(selection=["counts"])
dataset_empty = SpectrumDataset.create(e_reco=np.logspace(-1, 2, 30) * u.TeV,
region=rectangle)
datasets = []
for obs in observations:
dataset = dataset_maker.run(dataset_empty, obs)
dataset_on_off = bkg_maker.run(observation=obs, dataset=dataset)
datasets.append(dataset_on_off)
m = Map.create(skydir=crab_position, width=(8, 8), proj="TAN")
_, ax, _ = m.plot(vmin=-1, vmax=0)
rectangle.to_pixel(ax.wcs).plot(ax=ax, color="black")
plot_spectrum_datasets_off_regions(datasets=datasets, ax=ax)
plt.loglog()
print(cta_irf["aeff"].data)
# In[ ]:
edisp = cta_irf["edisp"].to_energy_dispersion(offset=offset,
e_true=energy,
e_reco=energy)
edisp.plot_matrix()
print(edisp.data)
# In[ ]:
dataset = SpectrumDataset(aeff=aeff,
edisp=edisp,
model=model_ref,
livetime=livetime,
obs_id=0)
dataset.fake(random_state=42)
# In[ ]:
# Take a quick look at the simulated counts
dataset.counts.plot()
# ## Include Background
#
# In this section we will include a background component. Furthermore, we will also simulate more than one observation and fit each one individually in order to get average fit results.
# In[ ]:
class TestSpectrumDataset:
"""Test fit on counts spectra without any IRFs"""
def setup(self):
self.nbins = 30
binning = np.logspace(-1, 1, self.nbins + 1) * u.TeV
self.source_model = SkyModel(spectral_model=PowerLawSpectralModel(
index=2.1,
amplitude=1e5 * u.Unit("cm-2 s-1 TeV-1"),
reference=0.1 * u.TeV,
))
self.livetime = 100 * u.s
aeff = EffectiveAreaTable.from_constant(binning, "1 cm2")
bkg_rate = np.ones(self.nbins) / u.s
bkg_expected = (bkg_rate * self.livetime).to_value("")
self.bkg = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=bkg_expected)
random_state = get_random_state(23)
flux = self.source_model.spectral_model.integral(
binning[:-1], binning[1:])
self.npred = (flux * aeff.data.data[0] * self.livetime).to_value("")
self.npred += bkg_expected
source_counts = random_state.poisson(self.npred)
self.src = CountsSpectrum(energy_lo=binning[:-1],
energy_hi=binning[1:],
data=source_counts)
self.dataset = SpectrumDataset(
models=self.source_model,
counts=self.src,
aeff=aeff,
livetime=self.livetime,
background=self.bkg,
name="test",
)
def test_data_shape(self):
assert self.dataset.data_shape[0] == self.nbins
def test_energy_range(self):
energy_range = self.dataset.energy_range
assert energy_range.unit == u.TeV
assert_allclose(energy_range.to_value("TeV"), [0.1, 10.0])
def test_cash(self):
"""Simple CASH fit to the on vector"""
fit = Fit([self.dataset])
result = fit.run()
# assert result.success
assert "minuit" in repr(result)
npred = self.dataset.npred().data.sum()
assert_allclose(npred, self.npred.sum(), rtol=1e-3)
assert_allclose(result.total_stat, -18087404.624, rtol=1e-3)
pars = result.parameters
assert_allclose(pars["index"].value, 2.1, rtol=1e-2)
assert_allclose(pars.error("index"), 0.00127, rtol=1e-2)
assert_allclose(pars["amplitude"].value, 1e5, rtol=1e-3)
assert_allclose(pars.error("amplitude"), 153.450, rtol=1e-2)
def test_fake(self):
"""Test the fake dataset"""
real_dataset = self.dataset.copy()
self.dataset.fake(314)
assert real_dataset.counts.data.shape == self.dataset.counts.data.shape
assert real_dataset.background.data.sum(
) == self.dataset.background.data.sum()
assert int(real_dataset.counts.data.sum()) == 907010
assert self.dataset.counts.data.sum() == 907331
def test_incorrect_mask(self):
mask_fit = np.ones(self.nbins, dtype=np.dtype("float"))
with pytest.raises(ValueError):
SpectrumDataset(
models=self.source_model,
counts=self.src,
livetime=self.livetime,
mask_fit=mask_fit,
background=self.bkg,
)
def test_set_model(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EDispKernel.from_diagonal_response(self.src.energy.edges,
self.src.energy.edges)
dataset = SpectrumDataset(None, self.src, self.livetime, None, aeff,
edisp, self.bkg)
spectral_model = PowerLawSpectralModel()
model = SkyModel(spectral_model=spectral_model, name="test")
dataset.models = model
assert dataset.models["test"] is model
models = Models([model])
dataset.models = models
assert dataset.models["test"] is model
def test_npred_models(self):
e_reco = MapAxis.from_energy_bounds("1 TeV", "10 TeV", nbin=3).edges
dataset = SpectrumDataset.create(e_reco=e_reco)
dataset.livetime = 1 * u.h
dataset.aeff.data.data += 1e10 * u.Unit("cm2")
pwl_1 = PowerLawSpectralModel(index=2)
pwl_2 = PowerLawSpectralModel(index=2)
model_1 = SkyModel(spectral_model=pwl_1)
model_2 = SkyModel(spectral_model=pwl_2)
dataset.models = Models([model_1, model_2])
npred = dataset.npred()
assert_allclose(npred.data.sum(), 64.8)
def test_str(self):
assert "SpectrumDataset" in str(self.dataset)
def test_spectrumdataset_create(self):
e_reco = u.Quantity([0.1, 1, 10.0], "TeV")
e_true = u.Quantity([0.05, 0.5, 5, 20.0], "TeV")
empty_dataset = SpectrumDataset.create(e_reco, e_true, name="test")
assert empty_dataset.name == "test"
assert empty_dataset.counts.total_counts == 0
assert empty_dataset.data_shape[0] == 2
assert empty_dataset.background.total_counts == 0
assert empty_dataset.background.energy.nbin == 2
assert empty_dataset.aeff.data.axis("energy").nbin == 3
assert empty_dataset.edisp.data.axis("e_reco").nbin == 2
assert empty_dataset.livetime.value == 0
assert len(empty_dataset.gti.table) == 0
assert empty_dataset.energy_range[0] is None
assert_allclose(empty_dataset.mask_safe, 0)
def test_spectrum_dataset_stack_diagonal_safe_mask(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp = EDispKernel.from_diagonal_response(self.src.energy.edges,
self.src.energy.edges)
livetime = self.livetime
dataset1 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime,
aeff=aeff,
edisp=edisp,
background=self.bkg.copy(),
)
livetime2 = 0.5 * livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:],
2 * aeff.data.data)
bkg2 = CountsSpectrum(
self.src.energy.edges[:-1],
self.src.energy.edges[1:],
data=2 * self.bkg.data,
)
safe_mask2 = np.ones_like(self.src.data, bool)
safe_mask2[0] = False
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp,
background=bkg2,
mask_safe=safe_mask2,
)
dataset1.stack(dataset2)
assert_allclose(dataset1.counts.data[1:], self.src.data[1:] * 2)
assert_allclose(dataset1.counts.data[0], self.src.data[0])
assert dataset1.livetime == 1.5 * self.livetime
assert_allclose(dataset1.background.data[1:], 3 * self.bkg.data[1:])
assert_allclose(dataset1.background.data[0], self.bkg.data[0])
assert_allclose(
dataset1.aeff.data.data.to_value("m2"),
4.0 / 3 * aeff.data.data.to_value("m2"),
)
assert_allclose(dataset1.edisp.pdf_matrix[1:], edisp.pdf_matrix[1:])
assert_allclose(dataset1.edisp.pdf_matrix[0],
0.5 * edisp.pdf_matrix[0])
def test_spectrum_dataset_stack_nondiagonal_no_bkg(self):
aeff = EffectiveAreaTable.from_parametrization(self.src.energy.edges,
"HESS")
edisp1 = EDispKernel.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.1, 0.0)
livetime = self.livetime
dataset1 = SpectrumDataset(counts=None,
livetime=livetime,
aeff=aeff,
edisp=edisp1,
background=None)
livetime2 = livetime
aeff2 = EffectiveAreaTable(self.src.energy.edges[:-1],
self.src.energy.edges[1:], aeff.data.data)
edisp2 = EDispKernel.from_gauss(self.src.energy.edges,
self.src.energy.edges, 0.2, 0.0)
dataset2 = SpectrumDataset(
counts=self.src.copy(),
livetime=livetime2,
aeff=aeff2,
edisp=edisp2,
background=None,
)
dataset1.stack(dataset2)
assert dataset1.counts is None
assert dataset1.background is None
assert dataset1.livetime == 2 * self.livetime
assert_allclose(dataset1.aeff.data.data.to_value("m2"),
aeff.data.data.to_value("m2"))
assert_allclose(dataset1.edisp.get_bias(1 * u.TeV), 0.0, atol=1.2e-3)
assert_allclose(dataset1.edisp.get_resolution(1 * u.TeV),
0.1581,
atol=1e-2)
def test_info_dict(self):
info_dict = self.dataset.info_dict()
assert_allclose(info_dict["n_on"], 907010)
assert_allclose(info_dict["background"], 3000.0)
assert_allclose(info_dict["significance"], 2924.522174)
assert_allclose(info_dict["excess"], 904010)
assert_allclose(info_dict["livetime"].value, 1e2)
assert info_dict["name"] == "test"
@requires_dependency("matplotlib")
def test_peek(self):
with mpl_plot_check():
self.dataset.peek()
self.dataset.edisp = None
with mpl_plot_check():
self.dataset.peek()
@requires_dependency("matplotlib")
def test_plot_fit(self):
with mpl_plot_check():
self.dataset.plot_fit()
# Load the IRFs
# In this simulation, we use the CTA-1DC irfs shipped with gammapy.
irfs = load_cta_irfs(
"$GAMMAPY_DATA/cta-1dc/caldb/data/cta/1dc/bcf/South_z20_50h/irf_file.fits")
# In[ ]:
obs = Observation.create(pointing=pointing, livetime=livetime, irfs=irfs)
print(obs)
# In[ ]:
# Make the SpectrumDataset
dataset_empty = SpectrumDataset.create(e_reco=energy_axis.edges,
e_true=energy_axis_true.edges,
region=on_region)
maker = SpectrumDatasetMaker(selection=["aeff", "edisp", "background"])
dataset = maker.run(dataset_empty, obs)
# In[ ]:
# Set the model on the dataset, and fake
dataset.model = model
dataset.fake(random_state=42)
print(dataset)
# You can see that backgound counts are now simulated
# ### OnOff analysis
#