def testShuffledDataframeRelativeToJackknife(self):
# Same as test above, but also testing that reordering the data doesn't
# change results, up to order.
df = pd.DataFrame({
"X": range(11),
"Y": np.concatenate((np.zeros(6), np.ones(5))),
"Z": np.concatenate((np.zeros(3), np.ones(8)))
})
metric = metrics.Distribution("X", ["Z"])
se_method = standard_errors.Jackknife()
output = core.Analyze(df.iloc[np.random.permutation(11)]).relative_to(
comparisons.AbsoluteDifference(
"Y",
0)).with_standard_errors(se_method).calculate(metric).run()
output = (output.reset_index().sort_values(by=["Y", "Z"]).set_index(
["Y", "Z"]))
correct = pd.DataFrame(
np.array([[-0.2, 0.18100283490], [0.2, 0.18100283490]]),
columns=[
"X Distribution Absolute Difference",
"X Distribution Absolute Difference Jackknife SE"
],
index=pd.MultiIndex(levels=[[1.], [0., 1.]],
labels=[[0, 0], [0, 1]],
names=["Y", "Z"]))
correct = (correct.reset_index().sort_values(by=["Y", "Z"]).set_index(
["Y", "Z"]))
self.assertTrue(
all(output.index == correct.index)
and all(output.columns == correct.columns)
and np.all(abs(output.values - correct.values) < 1e-10))
def testDataframeRelativeToJackknife(self):
df = pd.DataFrame({
"X": range(11),
"Y": np.concatenate((np.zeros(6), np.ones(5))),
"Z": np.concatenate((np.zeros(3), np.ones(8)))
})
metric = metrics.Distribution("X", ["Z"])
se_method = standard_errors.Jackknife()
output = core.Analyze(df).relative_to(
comparisons.AbsoluteDifference(
"Y",
0)).with_standard_errors(se_method).calculate(metric).run()
correct = pd.DataFrame(
np.array([[-0.2, 0.18100283490], [0.2, 0.18100283490]]),
columns=[
"X Distribution Absolute Difference",
"X Distribution Absolute Difference Jackknife SE"
],
index=pd.MultiIndex(levels=[[1.], [0., 1.]],
labels=[[0, 0], [0, 1]],
names=["Y", "Z"]))
self.assertTrue(
all(output.index == correct.index)
and all(output.columns == correct.columns)
and np.all(abs(output.values - correct.values) < 1e-10))
def testWeightedDistribution(self):
df = pd.DataFrame({"X": [1, 1, 1, 2, 2, 3, 4],
"Y": [1, 2, 0, 1, 1, 1, 1]})
weights = np.array([1, 7, 1, 1, 1, 1, 1])
metric = metrics.Distribution("X", ["Y"])
output = metric(df, weights)
correct = pd.DataFrame(
np.array([12 / 20., 7 / 20., 1 / 20.]),
columns=[""],
index=pd.Index([1, 2, 0], name="Y"))
self.assertTrue(output.equals(correct))
def testTwoDimensionalDistribution(self):
df = pd.DataFrame({"X": [1, 1, 1, 2, 2, 3, 4],
"Y": [1, 2, 0, 1, 1, 1, 1],
"Z": [1, 0, 0, 0, 0, 0, 0]})
weights = np.array([1, 1, 1, 1, 1, 1, 1])
metric = metrics.Distribution("X", ["Y", "Z"])
output = metric(df, weights)
correct = pd.DataFrame(
np.array([1 / 14., 1 / 14., 1 / 14., 11 / 14.]),
columns=[""],
index=pd.MultiIndex(levels=[[0, 1, 2], [0, 1]],
labels=[[1, 2, 0, 1], [1, 0, 0, 0]],
names=["Y", "Z"]))
self.assertTrue(output.equals(correct))
def testSplitDataframe(self):
df = pd.DataFrame({
"X": range(11),
"Y": np.concatenate((np.zeros(6), np.ones(5))),
"Z": np.concatenate((np.zeros(3), np.ones(8)))
})
metric = metrics.Distribution("X", ["Z"])
output = core.Analyze(df).split_by(["Y"]).calculate(metric).run()
correct = pd.DataFrame(np.array([0.2, 0.8, 0.0, 1.0]),
columns=["X Distribution"],
index=pd.MultiIndex(levels=[[0.0, 1.0],
[0.0, 1.0]],
labels=[[0, 0, 1, 1],
[0, 1, 0, 1]],
names=["Y", "Z"]))
self.assertTrue(
all(output.index == correct.index)
and all(output.columns == correct.columns)
and np.all(abs(output.values - correct.values) < 1e-10))
def testDataframeJackknife(self):
df = pd.DataFrame({
"X": range(11),
"Y": np.concatenate((np.zeros(6), np.ones(5))),
"Z": np.concatenate((np.zeros(3), np.ones(8)))
})
metric = metrics.Distribution("X", ["Z"])
se_method = standard_errors.Jackknife("Y")
output = core.Analyze(df).with_standard_errors(se_method).calculate(
metric).run()
correct = pd.DataFrame(
np.array([[3 / 55.,
np.sqrt(((3 / 15. - 0.1)**2 + 0.1**2) / 2.)],
[52 / 55.,
np.sqrt(((12 / 15. - 0.9)**2 + 0.1**2) / 2.)]]),
columns=("X Distribution", "X Distribution Jackknife SE"),
index=pd.Index([0., 1.], name="Z"))
self.assertTrue(
all(output.index == correct.index)
and all(output.columns == correct.columns)
and np.all(abs(output.values - correct.values) < 1e-10))