def test_no_limits(self):
P = JointNormal(
mu=[1, 1],
labels=['x', 'y'],
cov=[
[1, 1],
[1, 1.01],
], N=10)
with self.assertRaises(ValueError):
P.probability(x=2, y=2)
def test_correct_value(self):
P = JointNormal(
mu=[0, 0],
labels=['x', 'y'],
cov=[
[1, 0],
[0, 1],
], N=10)
N = stats.norm()
self.assertAlmostEqual(P.probability(x__gt=0, y=0), N.cdf(0), places=14)
self.assertAlmostEqual(P.probability(x__lt=1, y=0), N.cdf(1), places=14)
self.assertAlmostEqual(P.probability(x__gt=5, y=0), 1. - N.cdf(5), places=14)
def test_sums_to_one(self):
P = JointNormal(
mu=[1, 1],
labels=['x', 'y'],
cov=[
[1, 1],
[1, 1.01],
], N=10)
px_gt = P.probability(x__gt=2, y=2)
px_lt = P.probability(x__lt=2, y=2)
py_gt = P.probability(y__gt=2, x=-20)
py_lt = P.probability(y__lt=2, x=-20)
self.assertAlmostEqual(sum([px_gt, px_lt]), 1, places=14)
self.assertAlmostEqual(sum([py_gt, py_lt]), 1, places=14)
def test_iterable(self):
P = JointNormal(
mu=[0, 0],
labels=['x', 'y'],
cov=[
[1, 0],
[0, 1],
], N=10)
x0 = range(-5, 6)
p_gt = np.array(P.probability(x__gt=x0))
p_lt = np.array(P.probability(x__lt=x0, y=2))
p_sum = p_gt + p_lt
self.assertTrue(all(round(x, 2) == 1 for x in p_sum))
self.assertAlmostEqual(p_gt[5], .5)
self.assertTrue(p_gt[0] > .5)
self.assertTrue(p_lt[0] < .5)
self.assertTrue(p_gt[-1] < .5)
self.assertTrue(p_lt[-1] > .5)
def test_correct_value(self):
P = JointNormal(
mu=[0, 0],
labels=['x', 'y'],
cov=[
[1, 0],
[0, 1],
], N=10)
percentile_value = P.percentile(x=.75)
self.assertAlmostEqual(P.probability(x__lt=percentile_value), .75, places=12)
def test_iterable(self):
P = JointNormal(
mu=[0, 0],
labels=['x', 'y'],
cov=[
[1, 0],
[0, 1],
], N=10)
percentiles = [.25, .5, .75]
values_at_percentiles = P.percentile(x=percentiles)
new_percentiles = P.probability(x__lt=values_at_percentiles)
self.assertAlmostEqual(new_percentiles[0], percentiles[0])
self.assertAlmostEqual(new_percentiles[1], percentiles[1])
self.assertAlmostEqual(new_percentiles[2], percentiles[2])