def test_prob(self):
np.random.seed(992)
mean, sd = np.random.rand(2)
g = Gaussian(mean, sd)
norm = 1 / np.sqrt(2 * np.pi * sd**2)
self.assertTrue(np.allclose(g.prob(mean), norm))
self.assertTrue(np.allclose(g.prob(mean + sd), np.exp(-1 / 2) * norm))
def test_sample(self):
n_samples = 10000
np.random.seed(37)
mean, sd = np.random.rand(2)
g = Gaussian(mean, sd)
samples = g.sample(n_samples)
mean_sd = sd / np.sqrt(n_samples) # by central limit theorem
self.assertTrue(np.allclose(mean, samples.mean(), atol=3 * mean_sd))
self.assertTrue(np.allclose(sd, samples.std(ddof=1), atol=0.01))
def test_prob(self):
mean, sd = np.random.rand(2)
g = Gaussian(mean, sd)
bg = BoundedGaussian(mean, sd, -1, 2)
self.assertEqual(bg.prob(1), g.prob(1))
self.assertEqual(bg.prob(-1), g.prob(-1))
self.assertEqual(bg.prob(2), g.prob(2))
self.assertEqual(bg.prob(-2), 0)
self.assertEqual(bg.prob(3), 0)
def test_scale_factor():
p1 = Gaussian(3, 1)
assert_equal(p1.scale_factor, 3)
p2 = Gaussian(0, 2)
assert_equal(p2.scale_factor, 2)
p4 = Uniform(-1, 1, 0)
assert_equal(p4.scale_factor, 0.2)
p5 = Uniform(1, 4)
assert_equal(p5.scale_factor, 2.5)
p6 = Uniform(0, np.inf)
assert_equal(p6.scale_factor, 1)
assert_equal(p2.scale(10), 5)
assert_equal(p2.unscale(5), 10)
def test_generate_guess():
gold1 = np.array([[-0.091949, 0.270532], [-1.463350, 0.691041],
[1.081791, 0.220404], [-0.239325, 0.811950],
[-0.491129, 0.010526]])
gold2 = np.array([[-0.045974, 0.535266], [-0.731675, 0.745520],
[0.540895, 0.510202], [-0.119662, 0.805975],
[-0.245564, 0.405263]])
pars = [Gaussian(0, 1), Uniform(0, 1, 0.8)]
guess1 = generate_guess(pars, 5, seed=22)
guess2 = generate_guess(pars, 5, scaling=0.5, seed=22)
assert_allclose(guess1, gold1, atol=1e-5)
assert_allclose(guess2, gold2, atol=1e-5)
def test_basic(self):
expected = [Gaussian(0, 2), Gaussian(2, 2), Uniform(0, 40)]
evaluated = make_center_priors(self.image)
self.assertEqual(evaluated, expected)
def test_my_properties(self):
self.assertEqual(self.u, Uniform(1, 2))
self.assertEqual(self.g, Gaussian(1, 2))
self.assertEqual(self.b, BoundedGaussian(1, 2, 0, 3))
self.assertEqual(self.c, ComplexPrior(self.u, self.g))
def g(self):
return Gaussian(1, 2)
def test_gaussian():
g = Gaussian(1, 1)
assert_equal(g.guess, 1)
assert_obj_close(g.lnprob(0), gold_sigma)
def test_gaussian_constant_addition(self):
self.assertEqual(self.g + 1., Gaussian(2, 2.))
self.assertEqual(-self.g, Gaussian(-1, 2))
def test_guess(self):
mean = np.random.rand()
g = Gaussian(mean, 1)
self.assertEqual(g.guess, mean)
self.assertRaises(AttributeError, setattr, g, 'guess', 2) # property
def test_xy_uncertainty(self):
expected = [Gaussian(0, 4), Gaussian(2, 4), Uniform(0, 40)]
evaluated = make_center_priors(self.image, xy_uncertainty_pixels=2)
self.assertEqual(evaluated, expected)
def test_variance(self):
sd = np.random.rand()
g = Gaussian(0, sd)
self.assertEqual(g.variance, sd**2)
self.assertRaises(AttributeError, setattr, g, 'variance',
2) # property
def test_construction(self):
parameters = {'mu': 1, 'sd': 3, 'name': 'a'}
g = Gaussian(**parameters)
self.assertTrue(isinstance(g, Prior))
for key, val in parameters.items():
self.assertEqual(getattr(g, key), val)
def test_prior_array_math(self):
expected_sum = np.array([Gaussian(1, 2), Gaussian(2, 2)])
expected_product = np.array([Gaussian(1, 2), Gaussian(2, 4)])
self.assertTrue(np.all(self.g + np.array([0, 1]) == expected_sum))
self.assertTrue(np.all(self.g * np.array([1, 2]) == expected_product))
def test_adding_2_gaussians(self):
self.assertEqual(self.g + self.g, Gaussian(2, np.sqrt(8)))
diff_name_sum = Gaussian(1, 2, 'a') + Gaussian(1, 2, 'b')
self.assertEqual(diff_name_sum.name, 'GaussianSum')
def test_gaussian_multiplication(self):
self.assertEqual(2 * self.g, Gaussian(2, 4))
self.assertEqual(self.g * 2, Gaussian(2, 4))
self.assertEqual(self.g / 2, Gaussian(0.5, 1))
self.assertEqual(-1 * self.g, Gaussian(-1, 2))
self.assertEqual(self.g * (-1), Gaussian(-1, 2))
def test_z_range_units(self):
expected = [Gaussian(0, 2), Gaussian(2, 2), Uniform(2, 10)]
evaluated = make_center_priors(self.image, z_range_units=(2, 10))
self.assertEqual(evaluated, expected)
def test_lnprob(self):
g = Gaussian(0, 1)
self.assertTrue(np.allclose(g.lnprob(1), GOLD_SIGMA))
g = Gaussian(0, 2)
self.assertTrue(np.allclose(g.lnprob(2), GOLD_SIGMA - np.log(2)))
def test_prior_math():
u = Uniform(1, 2)
g = Gaussian(1, 2)
b = prior.BoundedGaussian(1, 2, 0, 3)
assert_equal(u + 1, Uniform(2, 3))
assert_equal(1 + u, Uniform(2, 3))
assert_equal(-u, Uniform(-2, -1))
assert_equal(1 - u, Uniform(-1, 0))
assert_equal(u - 1, Uniform(0, 1))
assert_equal(2 * u, Uniform(2, 4))
assert_equal(u * 2, Uniform(2, 4))
assert_equal(u / 2, Uniform(0.5, 1))
print(g + 1.)
assert_equal(g + 1., Gaussian(2, 2.))
assert_equal(-g, Gaussian(-1, 2))
assert_equal(b + 1., prior.BoundedGaussian(2., 2, 1., 4.))
assert_equal(-b, prior.BoundedGaussian(-1, 2, -3, 0))
assert_equal(2 * g, Gaussian(2, 4))
assert_equal(g * 2, Gaussian(2, 4))
assert_equal(g / 2, Gaussian(0.5, 1))
assert_equal(g + g, Gaussian(2, np.sqrt(8)))
assert_equal(g + np.array([0, 1]),
np.array([Gaussian(1, 2), Gaussian(2, 2)]))
assert_equal(g * np.array([1, 2]),
np.array([Gaussian(1, 2), Gaussian(2, 4)]))
with assert_raises(TypeError):
u + u
with assert_raises(TypeError):
g + b
with assert_raises(TypeError):
g + [0, 1]
with assert_raises(TypeError):
g * g
def test_gaussian():
g = Gaussian(1, 1)
assert_equal(g.guess, 1)
assert_obj_close(g.lnprob(0),gold_sigma)