def check_imgdata_convolved():
"""What is the behavior when we add the PSF to plot_pvalue?"""
r1 = ui.get_psf()
ui.plot_pvalue(c1, c1 + g1, conv_model=r1, num=40, bins=5)
tmp = ui.get_pvalue_results()
# these values are different to test_plot_pvalue_imgpsf_unconvolved
#
assert tmp.null == pytest.approx(2391.26963100235)
assert tmp.alt == pytest.approx(563.3992697080881)
assert tmp.lr == pytest.approx(1827.8703612942618)
assert tmp.samples.shape == (40, 1)
assert tmp.stats.shape == (40, 2)
assert tmp.ratios.shape == (40, )
tmp = ui.get_pvalue_plot()
assert tmp.lr == pytest.approx(1827.8703612942618)
assert tmp.xlabel == 'Likelihood Ratio'
assert tmp.ylabel == 'Frequency'
assert tmp.title == 'Likelihood Ratio Distribution'
assert tmp.ratios.shape == (40, )
assert tmp.xlo.shape == (6, )
assert tmp.xhi.shape == (6, )
assert tmp.y.shape == (6, )
def check_imgdata_unconvolved(caplog):
"""What is the behavior when we do not add the PSF to plot_pvalue?
Note we add a check of the screen output here.
"""
# include a check of the screen output
#
with caplog.at_level('INFO', logger='sherpa'):
with SherpaVerbosity('INFO'):
ui.plot_pvalue(c1, c1 + g1, num=40, bins=5)
assert len(caplog.records) == 1
lname, lvl, msg = caplog.record_tuples[0]
assert lname == 'sherpa.ui.utils'
assert lvl == logging.INFO
# Do not use equality tests for the numeric values in case
# there are numpy-version differences in the number of
# significant figures.
#
toks = msg.split('\n')
assert len(toks) == 5
assert toks[0] == 'Likelihood Ratio Test'
assert toks[1].startswith('null statistic = 2391.2696')
assert toks[2].startswith('alt statistic = 353.82')
assert toks[3].startswith('likelihood ratio = 2037.446')
assert toks[4] == 'p-value < 0.025'
tmp = ui.get_pvalue_results()
assert tmp.null == pytest.approx(2391.2696310023503)
assert tmp.alt == pytest.approx(353.8235336370698)
assert tmp.lr == pytest.approx(2037.4460973652804)
assert tmp.samples.shape == (40, 1)
assert tmp.stats.shape == (40, 2)
assert tmp.ratios.shape == (40, )
tmp = ui.get_pvalue_plot(2037.4460973652804)
assert tmp.lr == pytest.approx(2037.4460973652804)
assert tmp.xlabel == 'Likelihood Ratio'
assert tmp.ylabel == 'Frequency'
assert tmp.title == 'Likelihood Ratio Distribution'
assert tmp.ratios.shape == (40, )
assert tmp.xlo.shape == (6, )
assert tmp.xhi.shape == (6, )
assert tmp.y.shape == (6, )
def test_plot_pvalue(make_data_path, clean_astro_ui, hide_logging):
"""Check plot_pvalue with PHA data."""
fname = make_data_path("qso.pi")
ui.load_pha(fname)
ui.set_stat('cstat')
ui.set_method("neldermead")
ui.group_counts(10)
ui.notice(0.3, 8)
ui.set_model("xsphabs.abs1*(xspowerlaw.p1 +gauss1d.g1)")
# move the fit close to the best fit to save a small amount
# of time.
abs1.nh = 0.05
p1.phoindex = 1.28
p1.norm = 2e-4
g1.ampl = 1.8e-5
g1.pos = 3.
ui.freeze(g1.pos)
g1.fwhm = 0.1
ui.freeze(g1.fwhm)
# Could we reduce the number of bins to save evaluation time?
# We do want a non-default num value when checking the shapes
# of the output attributes.
#
ui.fit()
ui.plot_pvalue(p1, p1 + g1, num=100, bins=20)
tmp = ui.get_pvalue_results()
assert tmp.null == pytest.approx(210.34566845619273)
assert tmp.alt == pytest.approx(207.66618095925094)
assert tmp.lr == pytest.approx(2.679487496941789)
# Have we returned the correct info?
#
# Is it worth checking the stored data (aka how randomised is this
# output)?
#
assert tmp.samples.shape == (100, 2)
assert tmp.stats.shape == (100, 2)
assert tmp.ratios.shape == (100, )
# Check the plot
#
tmp = ui.get_pvalue_plot()
assert tmp.lr == pytest.approx(2.679487496941789)
assert tmp.xlabel == 'Likelihood Ratio'
assert tmp.ylabel == 'Frequency'
assert tmp.title == 'Likelihood Ratio Distribution'
# It would be nice to check the values here
#
assert tmp.ratios.shape == (100, )
assert tmp.xlo.shape == (21, )
assert tmp.xhi.shape == (21, )
assert tmp.y.shape == (21, )