def test_compare_sde_computations(n_samples=1000, n_bins=10):
y, pred, weights, bins, groups = generate_binned_dataset(n_samples=n_samples, n_bins=n_bins)
target_efficiencies = RandomState().uniform(size=3)
a = compute_sde_on_bins(pred[:, 1], mask=(y == 1), bin_indices=bins,
target_efficiencies=target_efficiencies, sample_weight=weights)
b = compute_sde_on_groups(pred[:, 1], mask=(y == 1), groups_indices=groups,
target_efficiencies=target_efficiencies, sample_weight=weights)
assert numpy.allclose(a, b)
def test_compare_sde_computations(n_samples=1000, n_bins=10):
y, pred, weights, bins, groups = generate_binned_dataset(
n_samples=n_samples, n_bins=n_bins)
target_efficiencies = RandomState().uniform(size=3)
a = compute_sde_on_bins(pred[:, 1],
mask=(y == 1),
bin_indices=bins,
target_efficiencies=target_efficiencies,
sample_weight=weights)
b = compute_sde_on_groups(pred[:, 1],
mask=(y == 1),
groups_indices=groups,
target_efficiencies=target_efficiencies,
sample_weight=weights)
assert numpy.allclose(a, b)
def test_cvm_sde_limit(size=2000):
""" Checks that in the limit CvM coincides with MSE """
effs = numpy.linspace(0, 1, 2000)
y_pred = random.uniform(size=size)
y = random.uniform(size=size) > 0.5
sample_weight = random.exponential(size=size)
bin_indices = random.randint(0, 10, size=size)
y_pred += bin_indices * random.uniform()
mask = y == 1
val1 = bin_based_cvm(y_pred[mask], sample_weight=sample_weight[mask], bin_indices=bin_indices[mask])
val2 = compute_sde_on_bins(y_pred, mask=mask, bin_indices=bin_indices, target_efficiencies=effs,
sample_weight=sample_weight)
assert numpy.allclose(val1, val2 ** 2, atol=1e-3, rtol=1e-2)
def test_cvm_sde_limit(size=2000):
""" Checks that in the limit CvM coincides with MSE """
effs = numpy.linspace(0, 1, 2000)
y_pred = random.uniform(size=size)
y = random.uniform(size=size) > 0.5
sample_weight = random.exponential(size=size)
bin_indices = random.randint(0, 10, size=size)
y_pred += bin_indices * random.uniform()
mask = y == 1
val1 = bin_based_cvm(y_pred[mask],
sample_weight=sample_weight[mask],
bin_indices=bin_indices[mask])
val2 = compute_sde_on_bins(y_pred,
mask=mask,
bin_indices=bin_indices,
target_efficiencies=effs,
sample_weight=sample_weight)
assert numpy.allclose(val1, val2**2, atol=1e-3, rtol=1e-2)
