def test_empirical_prior():
# set up some emprical prior
x = np.linspace(0, 40, 100) / 180 * np.pi
g1 = Gaussian1D(amplitude=.2, mean=21 / 180 * np.pi, stddev=0.05)
g1.bounding_box.amplitude = (0, 1.)
g1.bounding_box.mean = (0, np.pi / 2)
g2 = Gaussian1D(amplitude=0.05, mean=23 / 180 * np.pi, stddev=.05)
g2.bounding_box.amplitude = (.0, 1.)
g2.bounding_box.mean = (0, np.pi / 2)
# g is astropy CompoundModel
g = g1 + g2
# integration works
assert g(21 / 180 * np.pi) == pytest.approx(0.239186373422815)
assert empirical_prior(21 / 180 * np.pi, g) == np.log(0.239186373422815)
# 0. is included
assert g(0.) < 1e-12
assert g(0.) > 0.
assert empirical_prior(0., g) == np.log(g(0.))
# bounding box works
assert g(50. / 180. * np.pi) < 1e-12
assert g(50. / 180. * np.pi) > 0.
assert empirical_prior(50. / 180. * np.pi,
g) == np.log(g(50. / 180. * np.pi))
# outside of allowed inclination of pi/2
assert g(np.pi * .6) == pytest.approx(0.)
assert ~np.isfinite(empirical_prior(1.1 * np.pi, g))
# nan returns nan
assert np.isnan(g(np.nan))
assert np.isnan(empirical_prior(np.nan, g))
# cannot pass nan or float as g
with pytest.raises(TypeError) as e:
empirical_prior(0., np.nan)
with pytest.raises(TypeError) as e:
empirical_prior(0., 4.)
# g will accept a function like sqrt though
assert empirical_prior(1.2, np.sqrt) == pytest.approx(0.09116077839697724)
def test_inherit_constraints():
"""
Various tests for copying of constraint values between compound models and
their members.
Regression test for https://github.com/astropy/astropy/issues/3481
"""
model = (Gaussian1D(bounds={'stddev': (0, 0.3)}, fixed={'mean': True}) +
Gaussian1D(fixed={'mean': True}))
# Lots of assertions in this test as there are multiple interfaces to
# parameter constraints
assert 'stddev_0' in model.bounds
assert model.bounds['stddev_0'] == (0, 0.3)
assert model.stddev_0.bounds == (0, 0.3)
assert 'mean_0' in model.fixed
assert model.fixed['mean_0'] is True
assert model.mean_0.fixed is True
assert 'mean_1' in model.fixed
assert model.fixed['mean_1'] is True
assert model.mean_1.fixed is True
assert model.stddev_0 is model[0].stddev
# Great, all the constraints were inherited properly
# Now what about if we update them through the sub-models?
model.stddev_0.bounds = (0, 0.4)
assert model[0].stddev.bounds == (0, 0.4)
assert model[0].bounds['stddev'] == (0, 0.4)
model.stddev_0.bounds = (0.1, 0.5)
assert model[0].stddev.bounds == (0.1, 0.5)
assert model[0].bounds['stddev'] == (0.1, 0.5)
model[1].mean.fixed = False
assert model.mean_1.fixed is False
assert model[1].mean.fixed is False
# Now turn off syncing of constraints
assert model.bounds['stddev_0'] == (0.1, 0.5)
model.sync_constraints = False
model[0].stddev.bounds = (0, 0.2)
assert model.bounds['stddev_0'] == (0.1, 0.5)
model.sync_constraints = True
assert model.bounds['stddev_0'] == (0, 0.2)
def test_expression_formatting():
"""
Test that the expression strings from compound models are formatted
correctly.
"""
# For the purposes of this test it doesn't matter a great deal what
# model(s) are used in the expression, I don't think
G = Gaussian1D(1, 1, 1)
G2 = Gaussian2D(1, 2, 3, 4, 5, 6)
M = G + G
assert M._format_expression() == '[0] + [1]'
M = G + G + G
assert M._format_expression() == '[0] + [1] + [2]'
M = G + G * G
assert M._format_expression() == '[0] + [1] * [2]'
M = G * G + G
assert M._format_expression() == '[0] * [1] + [2]'
M = G + G * G + G
assert M._format_expression() == '[0] + [1] * [2] + [3]'
M = (G + G) * (G + G)
assert M._format_expression() == '([0] + [1]) * ([2] + [3])'
# This example uses parentheses in the expression, but those won't be
# preserved in the expression formatting since they technically aren't
# necessary, and there's no way to know that they were originally
# parenthesized (short of some deep, and probably not worthwhile
# introspection)
M = (G * G) + (G * G)
assert M._format_expression() == '[0] * [1] + [2] * [3]'
M = G**G
assert M._format_expression() == '[0] ** [1]'
M = G + G**G
assert M._format_expression() == '[0] + [1] ** [2]'
M = (G + G)**G
assert M._format_expression() == '([0] + [1]) ** [2]'
M = G + G | G
assert M._format_expression() == '[0] + [1] | [2]'
M = G + (G | G)
assert M._format_expression() == '[0] + ([1] | [2])'
M = G & G | G2
assert M._format_expression() == '[0] & [1] | [2]'
M = G & (G | G)
assert M._format_expression() == '[0] & ([1] | [2])'
def test_gaussian1d_n_models():
g = Gaussian1D(
amplitude=[1 * u.J, 2. * u.J],
mean=[1 * u.m, 5000 * u.AA],
stddev=[0.1 * u.m, 100 * u.AA],
n_models=2)
assert_quantity_allclose(g(1.01 * u.m), [0.99501248, 0.] * u.J)
assert_quantity_allclose(
g(u.Quantity([1.01 * u.m, 5010 * u.AA])), [0.99501248, 1.990025] * u.J)
def fit_data_with_gaussian(x_values, y_values, amplitude=1., mean=0, stddev=1.):
g_init = Gaussian1D(amplitude, mean, stddev)
lpost = PSDLogLikelihood(x_values, y_values, g_init)
parest = ParameterEstimation()
res = parest.fit(lpost, [amplitude, mean, stddev], neg=True)
opt_amplitude = res.p_opt[0]
opt_mean = res.p_opt[1]
opt_stddev = res.p_opt[2]
return opt_amplitude, opt_mean, opt_stddev
def _init_knotmodel(amp_init=0.01, v_init=-60.0):
"""Initialize model for knot: a single Gaussian"""
bounds = {
'amplitude': [0, None],
'stddev': [KNOT_WMIN, KNOT_WMAX],
'mean': [KNOT_VMIN, KNOT_VMAX]
}
knotmodel = Gaussian1D(amp_init, v_init, 5.0, bounds=bounds)
return knotmodel
def update(self):
HaB_gauss = Gaussian1D(amplitude=self.HaB_amp,
stddev=vel2wl(6562.8, self.HaB_vel) / 2.35,
mean=self.Ha_mean +
vel2wl(self.Ha_mean, self.Ha_shift))
HaB_fl = HaB_gauss(self.wl)
self.HaB_spec.set_ydata(gauss_fl)
HaN_gauss = Gaussian1D(
amplitude=self.HaN_gauss,
sttdev=vel2wl(self.Ha_mean, self.HaN_vel) / 2.35,
mean=self.Ha_mean + vel2wl(self.Ha_mean, self.Ha_shift))
HaN_fl = HaN_gauss(self.wl)
self.HaN_spec.set_ydata(HaN_fl)
comp = HaN_fl + HaB_fl
self.comp_spec.set_ydata(comp)
self.fig.canvas.draw()
def test_from_list_list():
g1 = Gaussian1D(1, 4.6, 0.2)
g2 = Gaussian1D(2.5, 5.5, 0.1)
g3 = Gaussian1D(-1.7, 8.2, 0.1)
x = np.linspace(0, 10, 200)
y = g1(x) + g2(x) + g3(x)
spectrum = Spectrum1D(flux=y * u.Jy, spectral_axis=x * u.um)
lines = find_lines_derivative(spectrum, flux_threshold=0.01)
spec_reg = SpectralRegion.from_line_list(lines)
expected = [(4.072864321608041 * u.um, 5.072864321608041 * u.um),
(4.977386934673367 * u.um, 5.977386934673367 * u.um),
(7.690954773869347 * u.um, 8.690954773869347 * u.um)]
for i, reg in enumerate(expected):
assert_quantity_allclose(reg, (spec_reg[i].lower, spec_reg[i].upper))
def test_parameter_add_units():
"""
On the other hand, if starting from a parameter with no units, we should be
able to add units since this is unambiguous.
"""
g = Gaussian1D(1, 3, 0.1)
g.amplitude = 2 * u.Jy
assert_quantity_allclose(g.amplitude, 2 * u.Jy)
def test_model_1D_kernel(self):
"""
Check Model1DKernel against Gaussian1Dkernel
"""
stddev = 5.
gauss = Gaussian1D(1. / np.sqrt(2 * np.pi * stddev**2), 0, stddev)
model_gauss_kernel = Model1DKernel(gauss, x_size=21)
gauss_kernel = Gaussian1DKernel(stddev, x_size=21)
assert_almost_equal(model_gauss_kernel.array, gauss_kernel.array,
decimal=12)
def test_quantity_call():
"""
Test that if constructed with Quanties models must be called with quantities.
"""
g = Gaussian1D(mean=3 * u.m, stddev=3 * u.cm, amplitude=3 * u.Jy)
g(10 * u.m)
with pytest.raises(u.UnitsError):
g(10)
def test_compound_evaluate_named_param(expr):
"""
Tests that compound evaluate function produces the same
result as the models with the operator applied
"""
x = np.linspace(-5, 5, 10)
p1 = np.array([1, 0, 0.2])
p2 = np.array([3, 0.5, 0.5])
model1 = Gaussian1D(2, 1, 5)
model2 = Gaussian1D(2, 1, 5)
compound = expr(model1, model2)
assert_array_equal(
compound.evaluate(
x, *p2, amplitude_0=p1[0], mean_0=p1[1], stddev_0=p1[2]
),
expr(model1.evaluate(x, *p1), model2.evaluate(x, *p2)),
)
def test_PoissonLikelihoodFitter():
model = Gaussian1D(amplitude=1000, mean=2, stddev=3)
x = np.arange(-10, 20, 0.1)
dx = 0.1 * np.ones_like(x)
np.random.seed(0)
y = np.random.poisson(dx * model(x))
fitter = PoissonLikelihoodFitter()
model = fitter(model, x, y, dx)
expected = [995.29239606, 1.99019548, 3.00869128]
assert_allclose(model.parameters, expected, rtol=1e-3)
def test_inherit_constraints():
"""
Various tests for copying of constraint values between compound models and
their members.
Regression test for https://github.com/astropy/astropy/issues/3481
"""
model = (Gaussian1D(bounds={'stddev': (0, 0.3)}, fixed={'mean': True}) +
Gaussian1D(fixed={'mean': True}))
# We have to copy the model before modifying it, otherwise the test fails
# if it is run twice in a row, because the state of the model instance
# would be preserved from one run to the next.
model = deepcopy(model)
model.map_parameters()
# Lots of assertions in this test as there are multiple interfaces to
# parameter constraints
assert 'stddev_0' in model.bounds
assert model.bounds['stddev_0'] == (0, 0.3)
assert model.stddev_0.bounds == (0, 0.3)
assert 'mean_0' in model.fixed
assert model.fixed['mean_0'] is True
assert model.mean_0.fixed is True
assert 'mean_1' in model.fixed
assert model.fixed['mean_1'] is True
assert model.mean_1.fixed is True
assert model.stddev_0 is model[0].stddev
# Great, all the constraints were inherited properly
# Now what about if we update them through the sub-models?
model.stddev_0.bounds = (0, 0.4)
assert model[0].stddev.bounds == (0, 0.4)
assert model[0].bounds['stddev'] == (0, 0.4)
model.stddev_0.bounds = (0.1, 0.5)
assert model[0].stddev.bounds == (0.1, 0.5)
assert model[0].bounds['stddev'] == (0.1, 0.5)
model[1].mean.fixed = False
assert model.mean_1.fixed is False
assert model[1].mean.fixed is False
def test_indexing_on_instance():
"""Test indexing on compound model instances."""
m = Gaussian1D(1, 0, 0.1) + Const1D(2)
assert isinstance(m[0], Gaussian1D)
assert isinstance(m[1], Const1D)
# assert isinstance(m['Gaussian1D'], Gaussian1D)
# assert isinstance(m['Const1D'], Const1D)
# Test parameter equivalence
assert m[0].amplitude == 1
assert m[0].mean == 0
assert m[0].stddev == 0.1
assert m[1].amplitude == 2
# Similar couple of tests, but now where the compound model was created
# from model instances
g = Gaussian1D(1, 2, 3, name='g')
p = Polynomial1D(2, name='p')
m = g + p
assert m[0].name == 'g'
assert m[1].name == 'p'
assert m['g'].name == 'g'
assert m['p'].name == 'p'
# Test negative indexing
assert isinstance(m[-1], Polynomial1D)
assert isinstance(m[-2], Gaussian1D)
with pytest.raises(IndexError):
m[42]
with pytest.raises(IndexError):
m['foobar']
# Test string slicing
A = Const1D(1.1, name='A')
B = Const1D(2.1, name='B')
C = Const1D(3.1, name='C')
M = A + B * C
assert_allclose(M['B':'C'](1), 6.510000000000001)
def test_n_submodels():
"""
Test that CompoundModel.n_submodels properly returns the number
of components.
"""
g2 = Gaussian1D() + Gaussian1D()
assert g2.n_submodels() == 2
g3 = g2 + Gaussian1D()
assert g3.n_submodels() == 3
g5 = g3 | g2
assert g5.n_submodels() == 5
g7 = g5 / g2
assert g7.n_submodels() == 7
# make sure it works as class method
p = Polynomial1D + Polynomial1D
assert p.n_submodels() == 2
def test_parameter_quantity():
"""
Basic tests for initializing general models (that do not require units)
with parameters that have units attached.
"""
g = Gaussian1D(1 * u.J, 1 * u.m, 0.1 * u.m)
assert g.amplitude.value == 1.0
assert g.amplitude.unit is u.J
assert g.mean.value == 1.0
assert g.mean.unit is u.m
assert g.stddev.value == 0.1
assert g.stddev.unit is u.m
def test_fitting_incompatible_units():
"""
Raise an error if the data and model have incompatible units
"""
g_init = Gaussian1D(amplitude=1. * u.Jy, mean=3 * u.m, stddev=2 * u.cm)
fit_g = fitting.LevMarLSQFitter()
with pytest.raises(UnitsError) as exc:
fit_g(g_init, [1, 2, 3] * u.Hz, [4, 5, 6] * u.Jy)
assert exc.value.args[0] == (
"'Hz' (frequency) and 'm' (length) are not convertible")
def test_indexing_on_instance():
"""Test indexing on compound model instances."""
M = Gaussian1D + Const1D
m = M(1, 0, 0.1, 2)
assert isinstance(m[0], Gaussian1D)
assert isinstance(m[1], Const1D)
assert isinstance(m['Gaussian1D'], Gaussian1D)
assert isinstance(m['Const1D'], Const1D)
# Test parameter equivalence
assert m[0].amplitude == 1 == m.amplitude_0
assert m[0].mean == 0 == m.mean_0
assert m[0].stddev == 0.1 == m.stddev_0
assert m[1].amplitude == 2 == m.amplitude_1
# Test that parameter value updates are symmetric between the compound
# model and the submodel returned by indexing
const = m[1]
m.amplitude_1 = 42
assert const.amplitude == 42
const.amplitude = 137
assert m.amplitude_1 == 137
# Similar couple of tests, but now where the compound model was created
# from model instances
g = Gaussian1D(1, 2, 3, name='g')
p = Polynomial1D(2, name='p')
m = g + p
assert m[0].name == 'g'
assert m[1].name == 'p'
assert m['g'].name == 'g'
assert m['p'].name == 'p'
poly = m[1]
m.c0_1 = 12345
assert poly.c0 == 12345
poly.c1 = 6789
assert m.c1_1 == 6789
# Ensure this did *not* modify the original models we used as templates
assert p.c0 == 0
assert p.c1 == 0
# Test negative indexing
assert isinstance(m[-1], Polynomial1D)
assert isinstance(m[-2], Gaussian1D)
with pytest.raises(IndexError):
m[42]
with pytest.raises(IndexError):
m['foobar']
def gauss_removal(img, mask, linspace, where='bkg'):
"""
An additional step to remove cosmic rays. This fits a Gaussian to
the background (or a skewed Gaussian to the orders) and masks data
points which are above a certain sigma.
Parameters
----------
img : np.ndarray
Single exposure image.
mask : np.ndarray
An approximate mask for the orders.
linspace : array
Sets the lower and upper bin bounds for the
pixel values. Should be of length = 2.
where : str, optional
Sets where the mask is covering. Default is `bkg`.
Other option is `order`.
Returns
-------
img : np.ndarray
The same input image, now masked for newly identified
outliers.
"""
n, bins = np.histogram((img * mask).flatten(),
bins=np.linspace(linspace[0], linspace[1], 100))
bincenters = (bins[1:] + bins[:-1]) / 2
if where == 'bkg':
g = Gaussian1D(mean=0, amplitude=100, stddev=10)
rmv = np.where(np.abs(bincenters) <= 5)[0]
elif where == 'order':
GaussianSkewed = custom_model(skewed_gaussian)
g = GaussianSkewed(eta=0, omega=20, alpha=4, scale=100)
rmv = np.where(np.abs(bincenters) == 0)[0]
# finds bin centers and removes bincenter = 0 (because this bin
# seems to be enormous and we don't want to skew the best-fit
bincenters, n = np.delete(bincenters, rmv), np.delete(n, rmv)
# fit the model to the histogram bins
fitter = LevMarLSQFitter()
gfit = fitter(g, bincenters, n)
if where == 'bkg':
xcr, ycr = np.where(np.abs(img * mask) >= gfit.mean + 2 * gfit.stddev)
elif where == 'order':
xcr, ycr = np.where(img * mask <= gfit.eta - 1 * gfit.omega)
# returns an image that is nan-masked
img[xcr, ycr] = np.nan
return img
def test_fittable_compound():
m = Identity(1) | Mapping((0, )) | Gaussian1D(1, 5, 4)
x = np.arange(10)
y_real = m(x)
dy = 0.005
with NumpyRNGContext(1234567):
n = np.random.normal(0., dy, x.shape)
y_noisy = y_real + n
pfit = LevMarLSQFitter()
new_model = pfit(m, x, y_noisy)
y_fit = new_model(x)
assert_allclose(y_fit, y_real, atol=dy)
def test_evaluate_with_quantities():
"""
Test evaluation of a single model with Quantity parameters that do
not explicitly require units.
"""
# We create two models here - one with quantities, and one without. The one
# without is used to create the reference values for comparison.
g = Gaussian1D(1, 1, 0.1)
gq = Gaussian1D(1 * u.J, 1 * u.m, 0.1 * u.m)
# We first check that calling the Gaussian with quantities returns the
# expected result
assert_quantity_allclose(gq(1 * u.m), g(1) * u.J)
# Units have to be specified for the Gaussian with quantities - if not, an
# error is raised
with pytest.raises(UnitsError) as exc:
gq(1)
assert exc.value.args[0] == (
"Gaussian1D: Units of input 'x', (dimensionless), could not be "
"converted to required input units of m (length)")
# However, zero is a special case
assert_quantity_allclose(gq(0), g(0) * u.J)
# We can also evaluate models with equivalent units
assert_allclose(gq(0.0005 * u.km).value, g(0.5))
# But not with incompatible units
with pytest.raises(UnitsError) as exc:
gq(3 * u.s)
assert exc.value.args[0] == (
"Gaussian1D: Units of input 'x', s (time), could not be "
"converted to required input units of m (length)")
# We also can't evaluate the model without quantities with a quantity
with pytest.raises(UnitsError) as exc:
g(3 * u.m)
def __init__(self, *args, file_path=None, file_loader=None, embedded=False,
dev=False, skip_splash=False, **kwargs):
super(Application, self).__init__(*args, **kwargs)
# Set application icon
self.setWindowIcon(QIcon(":/icons/specviz.icns"))
# Load local plugins
self.load_local_plugins()
# cache the line lists for speedier access
from .core import linelist
linelist.populate_linelists_cache()
# Show splash
if not skip_splash:
self._splash_dialog = SplashDialog(2000)
self._splash_dialog.exec()
# If specviz is not being embded in another application, go ahead and
# perform the normal gui setup procedure.
if not embedded:
# Cache a reference to the currently active window
self.current_workspace = self.add_workspace()
# Add an initially empty plot
self.current_workspace.add_plot_window()
# Set embed mode state
self.current_workspace.set_embedded(embedded)
if dev:
y = Gaussian1D(mean=50, stddev=10)(np.arange(100)) + np.random.sample(100) * 0.1
spec1 = Spectrum1D(flux=y * u.Jy,
spectral_axis=np.arange(100) * u.AA)
spec2 = Spectrum1D(flux=np.random.sample(100) * u.erg,
spectral_axis=np.arange(100) * u.Hz)
spec3 = Spectrum1D(flux=np.random.sample(100) * u.erg,
spectral_axis=np.arange(100) * u.Hz)
data_item = self.current_workspace.model.add_data(spec1, "Spectrum 1")
self.current_workspace.model.add_data(spec2, "Spectrum 2")
self.current_workspace.model.add_data(spec3, "Spectrum 3")
# Set the first item as selected
self.current_workspace.force_plot(data_item)
# If a file path has been given, automatically add data
if file_path is not None:
self.current_workspace.load_data(
file_path, file_loader, display=True)
def evaluate(self, position):
"""Evaluate model at a given position.
Parameters
----------
position : `~astropy.coordinates.SkyCoord`
Position on the sky.
"""
glon, glat = position.galactic.l, position.galactic.b
width = self.width(glon)
amplitude = self.peak_brightness(glon)
mean = self.peak_latitude(glon)
return Gaussian1D.evaluate(glat, amplitude=amplitude, mean=mean, stddev=width)
def test_fitting_missing_model_units():
"""
Proceed if the data has units but the model doesn't
"""
x, y = _fake_gaussian_data()
g_init = Gaussian1D(amplitude=1., mean=3, stddev=2)
fit_g = fitting.LevMarLSQFitter()
g = fit_g(g_init, x, y)
assert_quantity_allclose(g.amplitude, 3 * u.Jy, rtol=0.05)
assert_quantity_allclose(g.mean, 1.3 * u.m, rtol=0.05)
assert_quantity_allclose(g.stddev, 0.8 * u.m, rtol=0.05)
g_init = Gaussian1D(amplitude=1., mean=3 * u.m, stddev=2 * u.m)
fit_g = fitting.LevMarLSQFitter()
g = fit_g(g_init, x, y)
assert_quantity_allclose(g.amplitude, 3 * u.Jy, rtol=0.05)
assert_quantity_allclose(g.mean, 1.3 * u.m, rtol=0.05)
assert_quantity_allclose(g.stddev, 0.8 * u.m, rtol=0.05)
def evaluate(self, position):
"""Evaluate model at a given position.
Parameters
----------
position : `~astropy.coordinates.SkyCoord`
Position on the sky.
"""
glon, glat = position.galactic.l, position.galactic.b
width = self.width(glon)
amplitude = self.peak_brightness(glon)
mean = self.peak_latitude(glon)
return Gaussian1D.evaluate(glat, amplitude=amplitude, mean=mean, stddev=width)
def psf1d(x,
y,
amplitude1=1.0,
stddev1=1.0,
amplitude2=0.5,
stddev2=5.0,
amplitude3=0.1,
stddev3=10.,
mean=0.,
offset=0.):
if amplitude1 < 0:
amplitude1 = 1e12
if amplitude2 < 0:
amplitude2 = 1e12
if amplitude3 < 0:
amplitude3 = 1e12
g1 = Gaussian1D(amplitude=amplitude1, mean=mean, stddev=stddev1)
g2 = Gaussian1D(amplitude=amplitude2, mean=mean, stddev=stddev2)
g3 = Gaussian1D(amplitude=amplitude3, mean=mean, stddev=stddev3)
y1 = g1.evaluate(x, amplitude=amplitude1, mean=mean, stddev=stddev1)
y2 = g2.evaluate(x, amplitude=amplitude2, mean=mean, stddev=stddev2)
y3 = g3.evaluate(x, amplitude=amplitude3, mean=mean, stddev=stddev3)
return y1 + y2 + y3 + offset
def test_fitting_with_initial_values():
x, y = _fake_gaussian_data()
# Fit the data using a Gaussian with units
g_init = Gaussian1D(amplitude=1. * u.mJy, mean=3 * u.cm, stddev=2 * u.mm)
fit_g = fitting.LevMarLSQFitter()
g = fit_g(g_init, x, y)
# TODO: update actual numerical results once implemented, but these should
# be close to the values below.
assert_quantity_allclose(g.amplitude, 3 * u.Jy, rtol=0.05)
assert_quantity_allclose(g.mean, 1.3 * u.m, rtol=0.05)
assert_quantity_allclose(g.stddev, 0.8 * u.m, rtol=0.05)
def test_pickle_compound():
"""
Regression test for
https://github.com/astropy/astropy/issues/3867#issuecomment-114547228
"""
# Test pickling a compound model class
GG = Gaussian1D + Gaussian1D
GG2 = pickle.loads(pickle.dumps(GG))
assert GG.param_names == GG2.param_names
assert GG.__name__ == GG2.__name__
# Test that it works, or at least evaluates successfully
assert GG()(0.12345) == GG2()(0.12345)
# Test pickling a compound model instance
g1 = Gaussian1D(1.0, 0.0, 0.1)
g2 = Gaussian1D([2.0, 3.0], [0.0, 0.0], [0.2, 0.3])
m = g1 + g2
m2 = pickle.loads(pickle.dumps(m))
assert m.param_names == m2.param_names
assert m.__class__.__name__ == m2.__class__.__name__
assert np.all(m.parameters == m2.parameters)
assert np.all(m(0) == m2(0))
# Test pickling a concrete class
p = pickle.dumps(_TestPickleModel, protocol=0)
# Note: This is very dependent on the specific protocol, but the point of
# this test is that the "concrete" model is pickled in a very simple way
# that only specifies the module and class name, and is unpickled by
# re-importing the class from the module in which it was defined. This
# should still work for concrete subclasses of compound model classes that
# were dynamically generated through an expression
exp = b'castropy.modeling.tests.test_compound\n_TestPickleModel\np0\n.'
# When testing against the expected value we drop the memo length field
# at the end, which may differ between runs
assert p[:p.rfind(b'p')] == exp[:exp.rfind(b'p')]
assert pickle.loads(p) is _TestPickleModel
def gaussfit1d(y):
global click, ii
click = np.zeros((2,2))
ii=0
def on_key(event):
global click, ii
#print 'event.key=%s, event.x=%d, event.y=%d, event.xdata=%f, \
#event.ydata=%f'%(event.key, event.x, event.y, event.xdata, event.ydata)
click[ii,0] = event.xdata
click[ii,1] = event.ydata
ii = ii + 1
return
x = np.array(range(len(y)))
print('Press any key at two points for specifying the '+\
'fitting region and the baselevel.')
fig=plt.figure()
cid = fig.canvas.mpl_connect('key_press_event', on_key)
plt.plot(x,y)
plt.draw()
while ii != 2:
plt.pause(0.1)
x1 = int(np.min(click[:,0])+0.5)
x2 = int(np.max(click[:,0])+0.5)
m = (click[1,1]-click[0,1])/(click[1,0]-click[0,0])
yy = y - m * ( x - click[0,0]) - click[0,1]
# Fit the data using a Gaussian
g_init = Gaussian1D(amplitude=np.min(yy[x1:x2+1]), \
mean=(x1+x2)/2.0, stddev=1.)
fit_g = LevMarLSQFitter()
g = fit_g(g_init, x[x1:x2+1], yy[x1:x2+1])
#print(g)
plt.plot((x1,x2), (y[x1],y[x2]))
xx = np.linspace(x1,x2,100)
yy = g(xx) + m * (xx - click[0,0]) + click[0,1]
plt.plot(xx,yy)
plt.draw()
try:
plt.pause(-1)
except:
pass
return g
def test_parameter_lose_units():
"""
Check that parameters that have been set to a quantity that are then set to
a value with no units raise an exception. We do this because setting a
parameter to a value with no units is ambiguous if units were set before:
if a parameter is 1 * u.Jy and the parameter is then set to 4, does this mean
2 without units, or 2 * u.Jy?
"""
g = Gaussian1D(1 * u.Jy, 3, 0.1)
with pytest.raises(UnitsError) as exc:
g.amplitude = 2
assert exc.value.args[0] == ("The 'amplitude' parameter should be given as "
"a Quantity because it was originally "
"initialized as a Quantity")
def _eval_y(self, y, pars):
"""Evaluate Gaussian model at a given ``y`` position.
"""
return Gaussian1D.eval(y, **pars)
def evaluate(self, x, y, sigma, c0, c1, c2, c3):
self.polynomial.coef = np.array([c0, c1, c2, c3])
poly_eval = self.polynomial(x)
poly_eval = (poly_eval + (self.window[0] + 1)) * self.window_delta / 2.
return Gaussian1D.evaluate(y, 1., poly_eval, sigma)