def test_bin_to_group_indices(size=100, bins=10):
bin_indices = RandomState().randint(0, bins, size=size)
mask = RandomState().randint(0, 2, size=size) > 0.5
group_indices = bin_to_group_indices(bin_indices, mask=mask)
assert numpy.sum([len(group) for group in group_indices]) == numpy.sum(mask)
a = numpy.sort(numpy.concatenate(group_indices))
b = numpy.where(mask > 0.5)[0]
assert numpy.all(a == b), 'group indices are computed wrongly'
def test_bin_to_group_indices(size=100, bins=10):
bin_indices = RandomState().randint(0, bins, size=size)
mask = RandomState().randint(0, 2, size=size) > 0.5
group_indices = bin_to_group_indices(bin_indices, mask=mask)
assert numpy.sum([len(group) for group in group_indices]) == numpy.sum(mask)
a = numpy.sort(numpy.concatenate(group_indices))
b = numpy.where(mask > 0.5)[0]
assert numpy.all(a == b), 'group indices are computed wrongly'
def generate_binned_dataset(n_samples, n_bins):
"""useful function, generates dataset with bins, groups, random weights.
This is used to test correlation functions. """
random = RandomState()
y = random.uniform(size=n_samples) > 0.5
pred = random.uniform(size=(n_samples, 2))
weights = random.exponential(size=(n_samples,))
bins = random.randint(0, n_bins, n_samples)
groups = bin_to_group_indices(bin_indices=bins, mask=(y == 1))
return y, pred, weights, bins, groups
def test_cvm(size=1000):
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)
mask = y == 1
groups_indices = bin_to_group_indices(bin_indices=bin_indices, mask=mask)
cvm1 = bin_based_cvm(y_pred[mask], sample_weight=sample_weight[mask], bin_indices=bin_indices[mask])
cvm2 = group_based_cvm(y_pred, mask=mask, sample_weight=sample_weight, groups_indices=groups_indices)
assert numpy.allclose(cvm1, cvm2)
def test_groups_matrix(size=1000, bins=4):
bin_indices = RandomState().randint(0, bins, size=size)
mask = RandomState().randint(0, 2, size=size) > 0.5
n_signal_events = numpy.sum(mask)
group_indices = bin_to_group_indices(bin_indices, mask=mask)
assert numpy.sum([len(group) for group in group_indices]) == n_signal_events
group_matrix = group_indices_to_groups_matrix(group_indices, n_events=size)
assert group_matrix.sum() == n_signal_events
for event_id, (is_signal, bin) in enumerate(zip(mask, bin_indices)):
assert group_matrix[bin, event_id] == is_signal
def generate_binned_dataset(n_samples, n_bins):
"""useful function, generates dataset with bins, groups, random weights.
This is used to test correlation functions. """
random = RandomState()
y = random.uniform(size=n_samples) > 0.5
pred = random.uniform(size=(n_samples, 2))
weights = random.exponential(size=(n_samples,))
bins = random.randint(0, n_bins, n_samples)
groups = bin_to_group_indices(bin_indices=bins, mask=(y == 1))
return y, pred, weights, bins, groups
def test_cvm(size=1000):
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)
mask = y == 1
groups_indices = bin_to_group_indices(bin_indices=bin_indices, mask=mask)
cvm1 = bin_based_cvm(y_pred[mask], sample_weight=sample_weight[mask], bin_indices=bin_indices[mask])
cvm2 = group_based_cvm(y_pred, mask=mask, sample_weight=sample_weight, groups_indices=groups_indices)
assert numpy.allclose(cvm1, cvm2)
def _compute_groups_indices(self, X, y, label):
"""Returns a list, each element is events' indices in some group."""
label_mask = y == label
extended_bin_limits = []
for var in self.uniform_features:
extended_bin_limits.append(numpy.percentile(X[var][label_mask], numpy.linspace(0, 100, 2 * self.n_bins + 1)))
groups_indices = list()
for shift in [0, 1]:
bin_limits = []
for axis_limits in extended_bin_limits:
bin_limits.append(axis_limits[1 + shift:-1:2])
bin_indices = compute_bin_indices(X.ix[:, self.uniform_features].values, bin_limits=bin_limits)
groups_indices += list(bin_to_group_indices(bin_indices, mask=label_mask))
return groups_indices