def test_choice_consistency(self):
N = 1000
K = 100
samples = []
random_state = np.random.RandomState(1)
for _ in range(N):
# unnormalized probability distribution
dist = random_state.randint(10, 1000, size=K).astype(float)
samples.append(_utils.choice(dist, random_state.rand()))
samples2 = []
random_state = np.random.RandomState(1)
for _ in range(N):
# unnormalized probability distribution
dist = random_state.randint(10, 1000, size=K).astype(float)
samples2.append(_utils.choice(dist, random_state.rand()))
self.assertEqual(samples, samples2)
def test_choice(self):
r = 0.9851367976958045
probz = np.array([0.14800396, 0.29600989, 0.46001528])
z = _utils.choice(probz, r)
self.assertEqual(z, len(probz) - 1)
N = 10000
K = 100
samples = []
random_state = np.random.RandomState(1)
for _ in range(N):
# unnormalized probability distribution
dist = random_state.randint(10, 1000, size=K).astype(float)
samples.append(_utils.choice(dist, random_state.rand()))
self.assertGreaterEqual(max(samples), 0)
self.assertLessEqual(max(samples), K-1)
p = 1/K
bound = 3*np.sqrt(N*p*(1-p))
counts = np.bincount(samples)
for k in range(K):
self.assertLess(abs(counts[k] - N*p), bound)