def testUnitBootstrap(self):
# The bootstrap depends upon random values to work; thus, we'll
# only check that it's statistically close to a simulated value.
# We set the seed to avoid flaky tests; this test will fail with
# probability 0.05 otherwise.
# Note this is an equivalent problem to the testBootstrap case,
# we've just split some rows.
np.random.seed(12345)
x = []
y = []
for ii in range(1, 101):
for _ in range(3):
x.append(ii)
y.append(ii)
data = pd.DataFrame({"X": x, "Y": y})
metric = metrics.Mean("X")
se_method = standard_errors.Bootstrap(100, unit="Y")
output = core.Analyze(data).with_standard_errors(se_method).calculate(
metric).run()
bootstrap_se = output["mean(X) Bootstrap SE"].values[0]
simulation_se = 2.88
epsilon = 0.41 # Two standard errors based on simulation.
self.assertAlmostEqual(simulation_se, bootstrap_se, delta=epsilon)
def testBootstrap(self):
# The bootstrap depends upon random values to work; thus, we'll
# only check that it's statistically close to the theoretical
# value.
# We set the seed to avoid flaky tests; this test will fail with
# probability 0.05 otherwise.
np.random.seed(12345)
data = pd.DataFrame({"X": range(1, 101)})
metric = metrics.Mean("X")
se_method = standard_errors.Bootstrap(100)
output = core.Analyze(data).with_standard_errors(se_method).calculate(
metric).run()
bootstrap_se = output["mean(X) Bootstrap SE"].values[0]
# Parameters based on the following R simulation
# set.seed(12345)
# library(bootstrap)
# x <- 1:100
# estimates <- replicate(1000, sd(bootstrap(x, 100, mean)$thetastar))
# mean(estimates)
# sd(estimates)
simulation_se = 2.88
epsilon = 0.41 # Two standard errors based on simulation.
self.assertAlmostEqual(simulation_se, bootstrap_se, delta=epsilon)
def testMeanWithWeights(self):
df = pd.DataFrame({"X": [1, 2, 3, 4]})
weights = np.array([3, 2, 1, 1])
metric = metrics.Mean("X")
output = metric(df, weights)
correct = 2.0
self.assertEqual(output, correct)
def testMultipleCalculations(self):
data = pd.DataFrame({"X": [1, 2, 3, 4, 5]})
output = core.Analyze(data).calculate(
[metrics.Sum("X"), metrics.Mean("X")]).run()
correct = pd.DataFrame(np.array([[15, 3.0]]),
columns=["sum(X)", "mean(X)"])
correct[["sum(X)"]] = correct[["sum(X)"]].astype(int)
self.assertTrue(output.equals(correct))