def test_stats_calc_stat_wstat_pha_nobg():
"""wstat statistic fails with no background: PHA dataset"""
statobj = WStat()
data, model = setup_single_pha(False, False, background=False)
with pytest.raises(StatErr):
statobj.calc_stat(data, model)
def test_stats_calc_stat_wstat_diffbins():
"""wstat statistic fails when src/bg bin sizes do not match"""
statobj = WStat()
data, model = setup_single_pha(True, False, background=True)
# Tweak data to have one-less bin than the background. This
# used to be easy but with data validation we need to
# create a new object.
#
data2 = DataPHA("faked",
channel=data.channel[:-1],
counts=data.counts[:-1],
staterror=data.staterror[:-1],
grouping=data.grouping[:-1],
exposure=data.exposure,
backscal=data.backscal,
areascal=data.areascal)
# We might expect the ARF/RMF calls to fail if we add validation
# (to check the ARF/RMF is valid for the PHA dataset).
#
data2.set_arf(data.get_arf())
data2.set_rmf(data.get_rmf())
data2.set_background(data.get_background())
# There is no Sherpa error for this, which seems surprising
with pytest.raises(TypeError) as err:
statobj.calc_stat(data2, model)
assert str(
err.value) == "input array sizes do not match, data: 5 vs group: 4"
def test_stats_calc_stat_wstat_nobg(usestat, usesys):
"""wstat statistic fails with no background"""
statobj = WStat()
data, model = setup_single(usestat, usesys)
with pytest.raises(StatErr):
statobj.calc_stat(data, model)
def test_wstat(self):
fit = Fit(self.data, self.model, WStat(), LevMar())
results = fit.fit()
# On a local linux machine I have to bump the tolerance to
# 3e-4, but this isn't seen on Travis. The fit isn't
# "great", so it may be that the results are sensitive to
# numerical differences (e.g. as introduced with updated
# compilers).
# tol = 3e-4
tol = 1e-6 # TODO: investigate difference
self.compare_results(self._fit_wstat_results_bench, results, tol=tol)
def test_stats_calc_stat_wstat_diffbins():
"""wstat statistic fails when src/bg bin sizes do not match"""
statobj = WStat()
data, model = setup_single_pha(True, False, background=True)
# Tweak data to have one-less bin than the background
data.channel = data.channel[:-1]
data.counts = data.channel[:-1]
for attr in ['staterror', 'syserror', 'grouping', 'quality', 'backscal']:
val = getattr(data, attr)
if val is not None:
try:
setattr(data, attr, val[:-1])
except TypeError:
# assume a scalar, so leave be
pass
# There is no Sherpa error for this, which seems surprising
with pytest.raises(TypeError):
statobj.calc_stat(data, model)
def test_wstat(hide_logging, reset_xspec, setup):
fit = Fit(setup['data'], setup['model'], WStat(), LevMar())
results = fit.fit()
_fit_wstat_results_bench = {
'succeeded':
1,
'numpoints':
460,
'dof':
457,
'istatval':
21647.48285025895,
'statval':
472.6585709918982,
'parvals':
numpy.array([1.750204250228727, 5.47466040324842, -1.9983562007031974])
}
compare_results(_fit_wstat_results_bench, results, tol=2e-4)
def test_wstat_error(self):
data = self.bkg
data.notice(0.5, 7.0)
fit = Fit(data, self.model, WStat(), NelderMead())
self.assertRaises(StatErr, fit.fit)
def test_wstat(self):
fit = Fit(self.data, self.model, WStat(), NelderMead())
results = fit.fit()
self.compare_results(self._fit_wstat_results_bench, results)
def prepare_spectra(group, nH, add_gal, redshift):
pha = read_pha("core_spectrum.pi")
pha.set_analysis("energy")
pha.notice(0.5, 7.0)
tabs = ~pha.mask
pha.group_counts(group, tabStops=tabs)
x = pha.get_x()
x = pha.apply_filter(x, pha._middle)
y = pha.get_y(filter=True)
pha.set_analysis("energy")
model = xsphabs.abs1 * powlaw1d.srcp1
print("Fitting the spectrum")
zFlag = False
if (nH is not None) and (nH > 0.0):
if add_gal == 1:
model = xsphabs.gal * xszphabs.abs1 * powlaw1d.srcp
gal.nH = nH
freeze(gal.nH)
zFlag = True
else:
model = xsphabs.abs1 * powlaw1d.srcp1
abs1.nH = nH
freeze(abs1.nH)
else:
model = xszphabs.abs1 * powlaw1d.srcp1
zFlag = True
if zFlag is True and add_gal == 1:
# print('REDSHIFT',redshift)
abs1.redshift = redshift
freeze(abs1.redshift)
full_model = RSPModelPHA(pha.get_arf(), pha.get_rmf(), pha, pha.exposure * model)
print(full_model)
fit = Fit(pha, full_model, method=MonCar(), stat=WStat())
res = fit.fit()
print(res.format())
print(fit.est_errors())
# calculate the p-value for wstat
mplot2 = ModelPlot()
mplot2.prepare(pha, full_model)
miu = mplot2.y * pha.exposure * 0.0146
obs = y * pha.exposure * 0.0146
c, ce, cv = gof_cstat(miu, obs)
print(f"C0={c},C_e={ce},C_v={cv}")
zval = (fit.calc_stat() - ce) / np.sqrt(cv)
if zval > 0:
pval = special.erfc(zval / np.sqrt(2))
else:
pval = special.erf(abs(zval) / np.sqrt(2))
print(f"p-value for wstat = {pval}")
set_data(pha)
set_model(model)
save_chart_spectrum("core_flux_chart.dat", elow=0.5, ehigh=7.0)
# save_chart_spectrum("core_flux_chart.rdb",format='text/tsv', elow=0.5, ehigh=7.0)
save_spectrum_rdb("core_flux_chart.dat")
def test_wstat_error(hide_logging, reset_xspec, setup_bkg):
fit = Fit(setup_bkg['bkg'], setup_bkg['model'], WStat(), NelderMead())
with pytest.raises(StatErr):
fit.fit()
def prepare_spectra(nH: float,
group: int = 1,
add_gal: bool = False,
redshift: Optional[float] = None,
**kwargs) -> float:
"""
Fit the spectra using an absorbed powerlaw model using the Wstat statistic. The function also returns a p-value for the gof.
:param nH: The galactic absorption column density in units of 10^22 /cm3
:param group: The number of counts per energy bin
:param add_gal: Setting this to True would add an intrinsic abrosption column density along side the galactic one
:param redshift: The redshift to use in the fit. Only takes effect if add_gal is set to True
...
:return: Returns the p-value of the gof. The null hypothesis states that the model and the observation differ while alternate says that the model explains the data
"""
pha = read_pha("core_spectrum.pi")
pha.set_analysis("energy")
pha.notice(0.5, 7.0)
tabs = ~pha.mask
pha.group_counts(group, tabStops=tabs)
x = pha.get_x()
x = pha.apply_filter(x, pha._middle)
y = pha.get_y(filter=True)
pha.set_analysis("energy")
model = xsphabs.abs1 * powlaw1d.srcp1
print("Fitting the spectrum")
zFlag = False
if (nH is not None) and (nH > 0.0):
if add_gal == 1:
model = xsphabs.gal * xszphabs.abs1 * powlaw1d.srcp
gal.nH = nH
freeze(gal.nH)
zFlag = True
else:
model = xsphabs.abs1 * powlaw1d.srcp1
abs1.nH = nH
freeze(abs1.nH)
else:
model = xszphabs.abs1 * powlaw1d.srcp1
zFlag = True
if zFlag is True and add_gal == 1:
# print('REDSHIFT',redshift)
abs1.redshift = redshift
freeze(abs1.redshift)
full_model = RSPModelPHA(pha.get_arf(), pha.get_rmf(), pha,
pha.exposure * model)
print(full_model)
fit = Fit(pha, full_model, method=MonCar(), stat=WStat())
res = fit.fit()
print(res.format())
print(fit.est_errors())
# calculate the p-value for wstat
mplot2 = ModelPlot()
mplot2.prepare(pha, full_model)
miu = mplot2.y * pha.exposure * 0.0146
obs = y * pha.exposure * 0.0146
c, ce, cv = SpecUtils.estimate_gof_cstat(miu, obs)
#print(f"C0={c},C_e={ce},C_v={cv}")
zval = (fit.calc_stat() - ce) / np.sqrt(cv)
if zval > 0:
pval = special.erfc(zval / np.sqrt(2))
else:
pval = special.erf(abs(zval) / np.sqrt(2))
print(f"p-value for wstat = {pval}")
set_data(pha)
set_model(model)
save_chart_spectrum("core_flux_chart.dat", elow=0.5, ehigh=7.0)
# save_chart_spectrum("core_flux_chart.rdb",format='text/tsv', elow=0.5, ehigh=7.0)
SAOTraceUtils.save_spectrum_rdb("core_flux_chart.dat")
return pval
