def make_srcmap_old(psf,
spatial_model,
sigma,
npix=500,
xpix=0.0,
ypix=0.0,
cdelt=0.01,
rebin=1,
psf_scale_fn=None):
"""Compute the source map for a given spatial model.
Parameters
----------
psf : `~fermipy.irfs.PSFModel`
spatial_model : str
Spatial model.
sigma : float
Spatial size parameter for extended models.
xpix : float
Source position in pixel coordinates in X dimension.
ypix : float
Source position in pixel coordinates in Y dimension.
rebin : int
Factor by which the original map will be oversampled in the
spatial dimension when computing the model.
psf_scale_fn : callable
Function that evaluates the PSF scaling function.
Argument is energy in MeV.
"""
if rebin > 1:
npix = npix * rebin
xpix = xpix * rebin + (rebin - 1.0) / 2.
ypix = ypix * rebin + (rebin - 1.0) / 2.
cdelt = cdelt / rebin
if spatial_model == 'RadialGaussian':
k = utils.make_cgauss_kernel(psf, sigma, npix, cdelt, xpix, ypix,
psf_scale_fn)
elif spatial_model == 'RadialDisk':
k = utils.make_cdisk_kernel(psf, sigma, npix, cdelt, xpix, ypix,
psf_scale_fn)
elif spatial_model == 'PointSource':
k = utils.make_psf_kernel(psf, npix, cdelt, xpix, ypix, psf_scale_fn)
else:
raise Exception('Unsupported spatial model: %s', spatial_model)
if rebin > 1:
k = utils.sum_bins(k, 1, rebin)
k = utils.sum_bins(k, 2, rebin)
k *= psf.exp[:, np.newaxis, np.newaxis] * np.radians(cdelt)**2
return k
def calc_counts(skydir, ltc, event_class, event_types,
egy_bins, cth_bins, fn, npts=1):
"""Calculate the expected counts vs. true energy and incidence angle
for a source with spectral parameterization ``fn``.
Parameters
----------
skydir : `~astropy.coordinate.SkyCoord`
ltc : `~fermipy.irfs.LTCube`
egy_bins : `~numpy.ndarray`
Bin edges in observed energy in MeV.
cth_bins : `~numpy.ndarray`
Bin edges in cosine of the true incidence angle.
npts : int
Number of points by which to oversample each energy bin.
"""
#npts = int(np.ceil(32. / bins_per_dec(egy_bins)))
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
exp = calc_exp(skydir, ltc, event_class, event_types,
egy_bins, cth_bins)
dnde = fn.dnde(egy_bins)
cnts = loglog_quad(egy_bins, exp * dnde[:, None], 0)
cnts = sum_bins(cnts, 0, npts)
return cnts
def shift_to_coords(self, pix, fill_value=np.nan):
"""Create a new map that is shifted to the pixel coordinates
``pix``."""
pix_offset = self.get_offsets(pix)
dpix = np.zeros(len(self.shape) - 1)
for i in range(len(self.shape) - 1):
x = self.rebin * (pix[i] - pix_offset[i + 1]) + (self.rebin -
1.0) / 2.
dpix[i] = x - self._pix_ref[i]
pos = [
pix_offset[i] + self.shape[i] // 2 for i in range(self.data.ndim)
]
s0, s1 = utils.overlap_slices(self.shape_out, self.shape, pos)
k = np.zeros(self.data.shape)
for i in range(k.shape[0]):
k[i] = shift(self._data_spline[i],
dpix,
cval=np.nan,
order=2,
prefilter=False)
for i in range(1, len(self.shape)):
k = utils.sum_bins(k, i, self.rebin)
k0 = np.ones(self.shape_out) * fill_value
if k[s1].size == 0 or k0[s0].size == 0:
return k0
k0[s0] = k[s1]
return k0
def calc_counts(skydir,
ltc,
event_class,
event_types,
egy_bins,
cth_bins,
fn,
npts=1):
"""Calculate the expected counts vs. true energy and incidence angle
for a source with spectral parameterization ``fn``.
Parameters
----------
skydir : `~astropy.coordinate.SkyCoord`
ltc : `~fermipy.irfs.LTCube`
egy_bins : `~numpy.ndarray`
Bin edges in observed energy in MeV.
cth_bins : `~numpy.ndarray`
Bin edges in cosine of the true incidence angle.
npts : int
Number of points by which to oversample each energy bin.
"""
#npts = int(np.ceil(32. / bins_per_dec(egy_bins)))
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
exp = calc_exp(skydir, ltc, event_class, event_types, egy_bins, cth_bins)
dnde = fn.dnde(egy_bins)
cnts = loglog_quad(egy_bins, exp * dnde[:, None], 0)
cnts = sum_bins(cnts, 0, npts)
return cnts
def shift_to_coords(self, pix, fill_value=np.nan):
"""Create a new map that is shifted to the pixel coordinates
``pix``."""
pix_offset = self.get_offsets(pix)
dpix = np.zeros(len(self.shape) - 1)
for i in range(len(self.shape) - 1):
x = self.rebin * (pix[i] - pix_offset[i + 1]
) + (self.rebin - 1.0) / 2.
dpix[i] = x - self._pix_ref[i]
pos = [pix_offset[i] + self.shape[i] // 2
for i in range(self.data.ndim)]
s0, s1 = utils.overlap_slices(self.shape_out, self.shape, pos)
k = np.zeros(self.data.shape)
for i in range(k.shape[0]):
k[i] = shift(self._data_spline[i], dpix, cval=np.nan,
order=2, prefilter=False)
for i in range(1, len(self.shape)):
k = utils.sum_bins(k, i, self.rebin)
k0 = np.ones(self.shape_out) * fill_value
if k[s1].size == 0 or k0[s0].size == 0:
return k0
k0[s0] = k[s1]
return k0
def calc_counts_edisp(skydir, ltc, event_class, event_types,
egy_bins, cth_bins, fn, nbin=64, npts=1):
"""Calculate the expected counts vs. observed energy and true
incidence angle for a source with spectral parameterization ``fn``.
Parameters
----------
skydir : `~astropy.coordinate.SkyCoord`
ltc : `~fermipy.irfs.LTCube`
egy_bins : `~numpy.ndarray`
Bin edges in observed energy in MeV.
cth_bins : `~numpy.ndarray`
Bin edges in cosine of the true incidence angle.
npts : int
Number of points by which to oversample each energy bin.
"""
#npts = int(np.ceil(32. / bins_per_dec(egy_bins)))
# Split energy bins
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
etrue_bins = 10**np.linspace(1.0, 6.5, nbin * 5.5 + 1)
drm = calc_drm(skydir, ltc, event_class, event_types,
egy_bins, cth_bins, nbin=nbin)
cnts_etrue = calc_counts(skydir, ltc, event_class, event_types,
etrue_bins, cth_bins, fn)
cnts = np.sum(cnts_etrue[None, :, :] * drm[:, :, :], axis=1)
cnts = sum_bins(cnts, 0, npts)
return cnts
def make_srcmap_old(psf, spatial_model, sigma, npix=500, xpix=0.0, ypix=0.0,
cdelt=0.01, rebin=1, psf_scale_fn=None):
"""Compute the source map for a given spatial model.
Parameters
----------
psf : `~fermipy.irfs.PSFModel`
spatial_model : str
Spatial model.
sigma : float
Spatial size parameter for extended models.
xpix : float
Source position in pixel coordinates in X dimension.
ypix : float
Source position in pixel coordinates in Y dimension.
rebin : int
Factor by which the original map will be oversampled in the
spatial dimension when computing the model.
psf_scale_fn : callable
Function that evaluates the PSF scaling function.
Argument is energy in MeV.
"""
if rebin > 1:
npix = npix * rebin
xpix = xpix * rebin + (rebin - 1.0) / 2.
ypix = ypix * rebin + (rebin - 1.0) / 2.
cdelt = cdelt / rebin
if spatial_model == 'RadialGaussian':
k = utils.make_cgauss_kernel(psf, sigma, npix, cdelt,
xpix, ypix, psf_scale_fn)
elif spatial_model == 'RadialDisk':
k = utils.make_cdisk_kernel(psf, sigma, npix, cdelt,
xpix, ypix, psf_scale_fn)
elif spatial_model == 'PointSource':
k = utils.make_psf_kernel(psf, npix, cdelt,
xpix, ypix, psf_scale_fn)
else:
raise Exception('Unsupported spatial model: %s', spatial_model)
if rebin > 1:
k = utils.sum_bins(k, 1, rebin)
k = utils.sum_bins(k, 2, rebin)
k *= psf.exp[:, np.newaxis, np.newaxis] * np.radians(cdelt) ** 2
return k
def calc_counts_edisp(skydir,
ltc,
event_class,
event_types,
egy_bins,
cth_bins,
fn,
nbin=16,
npts=1):
"""Calculate the expected counts vs. observed energy and true
incidence angle for a source with spectral parameterization ``fn``.
Parameters
----------
skydir : `~astropy.coordinate.SkyCoord`
ltc : `~fermipy.irfs.LTCube`
egy_bins : `~numpy.ndarray`
Bin edges in observed energy in MeV.
cth_bins : `~numpy.ndarray`
Bin edges in cosine of the true incidence angle.
nbin : int
Number of points per decade with which to sample true energy.
npts : int
Number of points by which to oversample each reconstructed energy bin.
"""
#npts = int(np.ceil(32. / bins_per_dec(egy_bins)))
# Split energy bins
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
etrue_bins = 10**np.linspace(1.0, 6.5, np.int(np.ceil(nbin * 5.5 + 1)))
drm = calc_drm(skydir,
ltc,
event_class,
event_types,
egy_bins,
cth_bins,
nbin=nbin)
cnts_etrue = calc_counts(skydir, ltc, event_class, event_types, etrue_bins,
cth_bins, fn)
cnts = np.sum(cnts_etrue[None, :, :] * drm[:, :, :], axis=1)
cnts = sum_bins(cnts, 0, npts)
return cnts
def calc_drm(skydir, ltc, event_class, event_types,
egy_bins, cth_bins, nbin=64):
"""Calculate the detector response matrix."""
npts = int(np.ceil(128. / bins_per_dec(egy_bins)))
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
etrue_bins = 10**np.linspace(1.0, 6.5, nbin * 5.5 + 1)
egy = 10**utils.edge_to_center(np.log10(egy_bins))
egy_width = utils.edge_to_width(egy_bins)
etrue = 10**utils.edge_to_center(np.log10(etrue_bins))
edisp = create_avg_edisp(skydir, ltc, event_class, event_types,
egy, etrue, cth_bins)
edisp = edisp * egy_width[:, None, None]
edisp = sum_bins(edisp, 0, npts)
return edisp
def calc_drm(skydir,
ltc,
event_class,
event_types,
egy_bins,
cth_bins,
nbin=64):
"""Calculate the detector response matrix."""
npts = int(np.ceil(128. / bins_per_dec(egy_bins)))
egy_bins = np.exp(utils.split_bin_edges(np.log(egy_bins), npts))
etrue_bins = 10**np.linspace(1.0, 6.5, np.int(np.ceil(nbin * 5.5 + 1)))
egy = 10**utils.edge_to_center(np.log10(egy_bins))
egy_width = utils.edge_to_width(egy_bins)
etrue = 10**utils.edge_to_center(np.log10(etrue_bins))
edisp = create_avg_edisp(skydir, ltc, event_class, event_types, egy, etrue,
cth_bins)
edisp = edisp * egy_width[:, None, None]
edisp = sum_bins(edisp, 0, npts)
return edisp
def compute_norm(sig, bkg, ts_thresh, min_counts, sum_axes=None, bkg_fit=None,
rebin_axes=None):
"""Solve for the normalization of the signal distribution at which the
detection test statistic (twice delta-loglikelihood ratio) is >=
``ts_thresh`` AND the number of signal counts >= ``min_counts``.
This function uses the Asimov method to calculate the median
expected TS when the model for the background is fixed (no
uncertainty on the background amplitude).
Parameters
----------
sig : `~numpy.ndarray`
Array of signal amplitudes in counts.
bkg : `~numpy.ndarray`
Array of background amplitudes in counts.
ts_thresh : float
Test statistic threshold.
min_counts : float
Counts threshold.
sum_axes : list
Axes over which the source test statistic should be summed.
By default the summation will be performed over all
dimensions.
bkg_fit : `~numpy.ndarray`
Array of background amplitudes in counts for the fitting
model. If None then the fit model will be equal to the data
model.
"""
if sum_axes is None:
sum_axes = np.arange(sig.ndim)
sig = np.expand_dims(sig, -1)
bkg = np.expand_dims(bkg, -1)
sig_sum = np.apply_over_axes(np.sum, sig, sum_axes)
bkg_sum = np.apply_over_axes(np.sum, bkg, sum_axes)
bkg_fit_sum = None
if bkg_fit is not None:
bkg_fit = np.expand_dims(bkg_fit, -1)
bkg_fit_sum = np.apply_over_axes(np.sum, bkg_fit, sum_axes)
sig_rebin = sig
bkg_rebin = bkg
bkg_fit_rebin = bkg_fit
if rebin_axes:
sig_rebin = sig.copy()
bkg_rebin = bkg.copy()
if bkg_fit is not None:
bkg_fit_rebin = bkg_fit.copy()
for dim, rebin in zip(sum_axes, rebin_axes):
sig_rebin = sum_bins(sig_rebin, dim, rebin)
bkg_rebin = sum_bins(bkg_rebin, dim, rebin)
if bkg_fit is not None:
bkg_fit_rebin = sum_bins(bkg_fit_rebin, dim, rebin)
# Find approx solution using coarse binning and summed arrays
sig_scale = 10**np.linspace(0.0, 10.0, 51) * (min_counts / sig_sum)
vals_approx = _solve_norm(sig_rebin, bkg_rebin, ts_thresh, min_counts,
sig_scale, sum_axes, bkg_fit_rebin)
# Refine solution using an interval (0.1,10) around approx
# solution
sig_scale = (10**np.linspace(0.0, 1.0, 21) *
np.fmax(0.333 * vals_approx[..., None],
min_counts / sig_sum))
vals = _solve_norm(sig, bkg, ts_thresh, min_counts, sig_scale,
sum_axes, bkg_fit)
#sig_scale = 10**np.linspace(0.0, 10.0, 101)*(min_counts / sig_sum)
# vals = _solve_norm(sig, bkg, ts_thresh, min_counts, sig_scale2,
# sum_axes, bkg_fit)
return vals
def compute_norm(sig,
bkg,
ts_thresh,
min_counts,
sum_axes=None,
bkg_fit=None,
rebin_axes=None):
"""Solve for the normalization of the signal distribution at which the
detection test statistic (twice delta-loglikelihood ratio) is >=
``ts_thresh`` AND the number of signal counts >= ``min_counts``.
This function uses the Asimov method to calculate the median
expected TS when the model for the background is fixed (no
uncertainty on the background amplitude).
Parameters
----------
sig : `~numpy.ndarray`
Array of signal amplitudes in counts.
bkg : `~numpy.ndarray`
Array of background amplitudes in counts.
ts_thresh : float
Test statistic threshold.
min_counts : float
Counts threshold.
sum_axes : list
Axes over which the source test statistic should be summed.
By default the summation will be performed over all
dimensions.
bkg_fit : `~numpy.ndarray`
Array of background amplitudes in counts for the fitting
model. If None then the fit model will be equal to the data
model.
"""
if sum_axes is None:
sum_axes = np.arange(sig.ndim)
sig = np.expand_dims(sig, -1)
bkg = np.expand_dims(bkg, -1)
sig_sum = np.apply_over_axes(np.sum, sig, sum_axes)
bkg_sum = np.apply_over_axes(np.sum, bkg, sum_axes)
bkg_fit_sum = None
if bkg_fit is not None:
bkg_fit = np.expand_dims(bkg_fit, -1)
bkg_fit_sum = np.apply_over_axes(np.sum, bkg_fit, sum_axes)
sig_rebin = sig
bkg_rebin = bkg
bkg_fit_rebin = bkg_fit
if rebin_axes:
sig_rebin = sig.copy()
bkg_rebin = bkg.copy()
if bkg_fit is not None:
bkg_fit_rebin = bkg_fit.copy()
for dim, rebin in zip(sum_axes, rebin_axes):
sig_rebin = sum_bins(sig_rebin, dim, rebin)
bkg_rebin = sum_bins(bkg_rebin, dim, rebin)
if bkg_fit is not None:
bkg_fit_rebin = sum_bins(bkg_fit_rebin, dim, rebin)
# Find approx solution using coarse binning and summed arrays
sig_scale = 10**np.linspace(0.0, 10.0, 51) * (min_counts / sig_sum)
vals_approx = _solve_norm(sig_rebin, bkg_rebin, ts_thresh, min_counts,
sig_scale, sum_axes, bkg_fit_rebin)
# Refine solution using an interval (0.1,10) around approx
# solution
sig_scale = (10**np.linspace(0.0, 1.0, 21) *
np.fmax(0.333 * vals_approx[..., None], min_counts / sig_sum))
vals = _solve_norm(sig, bkg, ts_thresh, min_counts, sig_scale, sum_axes,
bkg_fit)
#sig_scale = 10**np.linspace(0.0, 10.0, 101)*(min_counts / sig_sum)
# vals = _solve_norm(sig, bkg, ts_thresh, min_counts, sig_scale2,
# sum_axes, bkg_fit)
return vals
