def test_compute_flux_spatial_no_psf():
    # check that spatial integration is not performed in the absence of a psf
    center = SkyCoord("0 deg", "0 deg", frame="galactic")
    region = CircleSkyRegion(center=center, radius=0.1 * u.deg)

    nbin = 2
    energy_axis_true = MapAxis.from_energy_bounds(".1 TeV",
                                                  "10 TeV",
                                                  nbin=nbin,
                                                  name="energy_true")

    spectral_model = ConstantSpectralModel()
    spatial_model = GaussianSpatialModel(lon_0=0 * u.deg,
                                         lat_0=0 * u.deg,
                                         frame="galactic",
                                         sigma="0.1 deg")

    models = SkyModel(spectral_model=spectral_model,
                      spatial_model=spatial_model)
    model = Models(models)

    exposure_region = RegionNDMap.create(region, axes=[energy_axis_true])
    exposure_region.data += 1.0
    exposure_region.unit = "m2 s"

    evaluator = MapEvaluator(model=model[0], exposure=exposure_region)
    flux = evaluator.compute_flux_spatial()

    assert_allclose(flux, 1.0)
def test_compute_flux_spatial():
    center = SkyCoord("0 deg", "0 deg", frame="galactic")
    region = CircleSkyRegion(center=center, radius=0.1 * u.deg)

    nbin = 2
    energy_axis_true = MapAxis.from_energy_bounds(".1 TeV",
                                                  "10 TeV",
                                                  nbin=nbin,
                                                  name="energy_true")

    spectral_model = ConstantSpectralModel()
    spatial_model = PointSpatialModel(lon_0=0 * u.deg,
                                      lat_0=0 * u.deg,
                                      frame="galactic")

    models = SkyModel(spectral_model=spectral_model,
                      spatial_model=spatial_model)
    model = Models(models)

    exposure_region = RegionNDMap.create(region,
                                         axes=[energy_axis_true],
                                         binsz_wcs="0.01deg")
    exposure_region.data += 1.0
    exposure_region.unit = "m2 s"

    geom = RegionGeom(region, axes=[energy_axis_true], binsz_wcs="0.01deg")
    psf = PSFKernel.from_gauss(geom.to_wcs_geom(), sigma="0.1 deg")

    evaluator = MapEvaluator(model=model[0], exposure=exposure_region, psf=psf)
    flux = evaluator.compute_flux_spatial()

    g = Gauss2DPDF(0.1)
    reference = g.containment_fraction(0.1)
    assert_allclose(flux.value, reference, rtol=0.003)
Exemple #3
0
def test_large_oversampling():
    nbin = 2
    energy_axis_true = MapAxis.from_energy_bounds(".1 TeV",
                                                  "10 TeV",
                                                  nbin=nbin,
                                                  name="energy_true")
    geom = WcsGeom.create(width=1, binsz=0.02, axes=[energy_axis_true])

    spectral_model = ConstantSpectralModel()
    spatial_model = GaussianSpatialModel(lon_0=0 * u.deg,
                                         lat_0=0 * u.deg,
                                         sigma=1e-4 * u.deg,
                                         frame="icrs")

    models = SkyModel(spectral_model=spectral_model,
                      spatial_model=spatial_model)
    model = Models(models)

    exposure = Map.from_geom(geom, unit="m2 s")
    exposure.data += 1.0

    psf = PSFKernel.from_gauss(geom, sigma="0.1 deg")

    evaluator = MapEvaluator(model=model[0], exposure=exposure, psf=psf)
    flux_1 = evaluator.compute_flux_spatial()

    spatial_model.sigma.value = 0.001
    flux_2 = evaluator.compute_flux_spatial()

    spatial_model.sigma.value = 0.01
    flux_3 = evaluator.compute_flux_spatial()

    spatial_model.sigma.value = 0.03
    flux_4 = evaluator.compute_flux_spatial()

    assert_allclose(flux_1.data.sum(), nbin, rtol=1e-4)
    assert_allclose(flux_2.data.sum(), nbin, rtol=1e-4)
    assert_allclose(flux_3.data.sum(), nbin, rtol=1e-4)
    assert_allclose(flux_4.data.sum(), nbin, rtol=1e-4)
Exemple #4
0
def test_compute_npred_sign():
    center = SkyCoord("0 deg", "0 deg", frame="galactic")
    energy_axis_true = MapAxis.from_energy_bounds(".1 TeV",
                                                  "10 TeV",
                                                  nbin=2,
                                                  name="energy_true")
    geom = WcsGeom.create(skydir=center,
                          width=1 * u.deg,
                          axes=[energy_axis_true],
                          frame='galactic',
                          binsz=0.2 * u.deg)

    spectral_model_pos = PowerLawSpectralModel(index=2,
                                               amplitude="1e-11 TeV-1 s-1 m-2")
    spectral_model_neg = PowerLawSpectralModel(
        index=2, amplitude="-1e-11 TeV-1 s-1 m-2")

    spatial_model = PointSpatialModel(lon_0=0 * u.deg,
                                      lat_0=0 * u.deg,
                                      frame="galactic")
    model_pos = SkyModel(spectral_model=spectral_model_pos,
                         spatial_model=spatial_model)
    model_neg = SkyModel(spectral_model=spectral_model_neg,
                         spatial_model=spatial_model)

    exposure = Map.from_geom(geom, unit="m2 s")
    exposure.data += 1.0

    psf = PSFKernel.from_gauss(geom, sigma="0.1 deg")

    evaluator_pos = MapEvaluator(model=model_pos, exposure=exposure, psf=psf)
    evaluator_neg = MapEvaluator(model=model_neg, exposure=exposure, psf=psf)

    npred_pos = evaluator_pos.compute_npred()
    npred_neg = evaluator_neg.compute_npred()

    assert (npred_pos.data == -npred_neg.data).all()
    assert np.all(npred_pos.data >= 0)
    assert np.all(npred_neg.data <= 0)