def test_function():
g = SkewNormal()
g.A.value = 2.5
g.x0.value = 0
g.scale.value = 1
g.shape.value = 0
assert_allclose(g.function(0), 1, rtol=3e-3)
assert_allclose(g.function(6), 1.52e-8, rtol=1e-3)
g.A.value = 5
g.x0.value = 4
g.scale.value = 3
g.shape.value = 2
assert_allclose(g.function(0), 6.28e-3, rtol=1e-3)
assert_allclose(g.function(g.mean), 2.855, rtol=1e-3)
def test_estimate_parameters_binned(only_current, binned, lazy, uniform):
s = Signal1D(np.empty((300,)))
s.axes_manager.signal_axes[0].is_binned = binned
axis = s.axes_manager.signal_axes[0]
axis.scale = 0.2
axis.offset = -10
g1 = SkewNormal(A=2, x0=2, scale=10, shape=5)
s.data = g1.function(axis.axis)
if not uniform:
axis.convert_to_non_uniform_axis()
if lazy:
s = s.as_lazy()
g2 = SkewNormal()
if binned and uniform:
factor = axis.scale
elif binned:
factor = np.gradient(axis.axis)
else:
factor = 1
assert g2.estimate_parameters(s, axis.low_value, axis.high_value,
only_current=only_current)
assert g2._axes_manager[-1].is_binned == binned
np.testing.assert_allclose(g1.A.value, g2.A.value * factor)
assert abs(g2.x0.value - g1.x0.value) <= 0.002
assert abs(g2.shape.value - g1.shape.value) <= 0.01
assert abs(g2.scale.value - g1.scale.value) <= 0.01
def test_function_nd(binned, lazy):
s = Signal1D(np.empty((300,)))
axis = s.axes_manager.signal_axes[0]
axis.scale = 0.2
axis.offset = -10
g1 = SkewNormal(A=2, x0=2, scale=10, shape=5)
s.data = g1.function(axis.axis)
s.metadata.Signal.binned = binned
s2 = stack([s] * 2)
if lazy:
s2 = s2.as_lazy()
g2 = SkewNormal()
factor = axis.scale if binned else 1
assert g2.estimate_parameters(s2, axis.low_value, axis.high_value, False)
assert g2.binned == binned
assert_allclose(g2.function_nd(axis.axis) * factor, s2.data, 0.06)
def test_estimate_parameters_binned(only_current, binned, lazy):
s = Signal1D(np.empty((300,)))
s.metadata.Signal.binned = binned
axis = s.axes_manager.signal_axes[0]
axis.scale = 0.2
axis.offset = -10
g1 = SkewNormal(A=2, x0=2, scale=10, shape=5)
s.data = g1.function(axis.axis)
if lazy:
s = s.as_lazy()
g2 = SkewNormal()
factor = axis.scale if binned else 1
assert g2.estimate_parameters(s, axis.low_value, axis.high_value,
only_current=only_current)
assert g2.binned == binned
assert_allclose(g1.A.value, g2.A.value * factor)
assert abs(g2.x0.value - g1.x0.value) <= 0.002
assert abs(g2.shape.value - g1.shape.value) <= 0.01
assert abs(g2.scale.value - g1.scale.value) <= 0.01
def test_fit(A=1, x0=0, shape=1, scale=1, noise=0.01):
"""
Creates a simulated noisy skew normal distribution based on the input
parameters and fits a skew normal component to this data.
"""
# create skew normal signal and add noise
g = SkewNormal(A=A, x0=x0, scale=scale, shape=shape)
x = np.arange(x0 - scale * 3, x0 + scale * 3, step=0.01 * scale)
s = Signal1D(g.function(x))
s.axes_manager.signal_axes[0].axis = x
s.add_gaussian_noise(std=noise * A)
# fit skew normal component to signal
g2 = SkewNormal()
m = s.create_model()
m.append(g2)
g2.x0.bmin = x0 - scale * 3 # prevent parameters to run away
g2.x0.bmax = x0 + scale * 3
g2.x0.bounded = True
m.fit(bounded=True)
m.print_current_values() # print out parameter values
m.plot() # plot fit
return m