def _(samples, estimator, repetitions=1000, probs=0.68):
estimand = torch.zeros(repetitions)
for i in range(repetitions):
bootstrap_values = resample(samples,
num_samples=len(samples),
replacement=True)
estimand[i] = estimator(bootstrap_values)
return pi(estimand, probs)
def _(dist, estimator, preprocess=None, repetitions=1000, probs=0.68):
estimand = torch.zeros(repetitions)
for i in range(repetitions):
samples = dist.sample()
if preprocess is not None:
samples = preprocess(samples)
# bootstrap_values = resample(samples, num_samples=len(samples), replacement=True)
bootstrap_values = np.random.choice(samples,
size=len(samples),
replace=True)
estimand[i] = estimator(bootstrap_values)
return pi(estimand, probs)
def get_percentile_confidence_interval(samples, probs):
pi = stats.pi(samples, prob=probs)
below_interval_1 = (samples < pi[0].item()).sum().float() / len(samples)
below_interval_2 = (samples < pi[1].item()).sum().float() / len(samples)
df = {
"LPI":
[round(below_interval_1.item(), 2) * 100,
round(pi[0].item(), 2)],
"UPI":
[round(below_interval_2.item(), 2) * 100,
round(pi[1].item(), 2)]
}
df = pd.DataFrame.from_dict(df)
return df
def _pi(x, dim=0):
return pi(x, prob=0.8, dim=dim)
def test_hpdi():
x = torch.randn(20000)
assert_equal(hpdi(x, prob=0.8), pi(x, prob=0.8), prec=0.01)
x = torch.empty(20000).exponential_(1)
assert_equal(hpdi(x, prob=0.2), torch.tensor([0.0, 0.22]), prec=0.01)
def test_pi():
x = torch.randn(1000).exp()
assert_equal(pi(x, prob=0.8), quantile(x, probs=[0.1, 0.9]))
def _(dist, estimator, repetitions=1000, probs=0.68):
samples = dist.sample((repetitions, ))
estimand = torch.zeros(repetitions)
for i in range(repetitions):
estimand[i] = estimator(samples[i])
return pi(estimand, probs)
def test_pi():
x = torch.empty(1000).log_normal_(0, 1)
assert_equal(pi(x, prob=0.8), quantile(x, probs=[0.1, 0.9]))
