def test_get_bootstrap_samples_multistat(spark_context, stack_depth=0):
data = np.concatenate([np.zeros(10000), np.ones(10000)])
res = mabsbb.get_bootstrap_samples(
spark_context,
data,
lambda x, y: {
'min': np.min(x),
'max': np.max(x),
'mean': np.dot(x, y),
},
num_samples=2
)
assert res.shape == (2, 3)
assert (res['min'] == 0).all()
assert (res['max'] == 1).all()
assert res['mean'].iloc[0] == pytest.approx(np.mean(data), rel=1e-1)
assert res['mean'].iloc[1] == pytest.approx(np.mean(data), rel=1e-1)
# If we stuff up (duplicate) the seeds then things aren't random
assert res['mean'].iloc[0] != res['mean'].iloc[1]
if stack_depth >= 3:
assert (res['mean'] != np.mean(data)).any() # Extremely unlikely
elif (res['mean'] == np.mean(data)).any():
# Re-roll the dice a few times to make sure this was a fluke.
test_get_bootstrap_samples_multistat(spark_context, stack_depth + 1)
def test_get_bootstrap_samples(spark_context):
res = mabsbb.get_bootstrap_samples(
spark_context, np.array([3., 3., 3.]), num_samples=2
)
assert res.shape == (2,)
assert res[0] == pytest.approx(3.)
assert res[1] == pytest.approx(3.)