def testFindChangePoints(self):
# Simple test that the output is as expected for a clear change.
# Tests for specific aspects of the algorithm are below.
data = [1, 1, 2, 1, 1, 8, 8, 8, 9, 8, 9]
series = list(enumerate(data))
expected = [
find_change_points.ChangePoint(
x_value=5,
median_before=1,
median_after=8,
window_start=1,
window_end=10,
size_before=4,
size_after=6,
relative_change=7,
std_dev_before=0.4330127018922193,
t_statistic=-24.452628375754593,
degrees_of_freedom=6.9938793160801023,
p_value=0.001)
]
actual = find_change_points.FindChangePoints(series,
max_window_size=10,
multiple_of_std_dev=3,
min_relative_change=0.5,
min_absolute_change=1,
min_steppiness=0.4,
min_segment_size=3)
self.assertEqual(expected, actual)
def SimulateAlertProcessing(chart_series, **config_dict):
"""Finds the same alerts as would be found normally as points are added.
Each time a new point is added to a data series on dashboard, the
FindChangePoints function is called with some points from that series.
In order to simulate this here, we need to repeatedly call FindChangePoints.
Args:
chart_series: A list of (x, y) pairs.
**config_dict: An alert threshold config dict.
Returns:
A list of find_change_points.ChangePoint objects, one for each alert found.
"""
all_change_points = []
highest_x = None # This is used to avoid finding duplicate alerts.
# The number of points that are passed in to FindChangePoints normally may
# depend on either the specific "max_window_size" value or another default
# used in find_anomalies.
window = config_dict.get('max_window_size', find_anomalies.DEFAULT_NUM_POINTS)
for end in range(1, len(chart_series)):
start = max(0, end - window)
series = chart_series[start:end]
change_points = find_change_points.FindChangePoints(series, **config_dict)
change_points = [c for c in change_points if c.x_value > highest_x]
if change_points:
highest_x = max(c.x_value for c in change_points)
all_change_points.extend(change_points)
return all_change_points
def FindChangePointsForTest(rows, config_dict):
"""Gets the anomaly data from the anomaly detection module.
Args:
rows: The Row entities to find anomalies for, sorted backwards by revision.
config_dict: Anomaly threshold parameters as a dictionary.
Returns:
A list of find_change_points.ChangePoint objects.
"""
data_series = [(revision, value) for (revision, _, value) in rows]
return find_change_points.FindChangePoints(data_series, **config_dict)
def _AssertFindsChangePoints(
self, y_values, expected_indexes, max_window_size=50, min_segment_size=6,
min_absolute_change=0, min_relative_change=0.01, min_steppiness=0.4,
multiple_of_std_dev=2.5):
"""Asserts that change points are found at particular indexes."""
series = list(enumerate(y_values))
results = find_change_points.FindChangePoints(
series,
max_window_size=max_window_size,
min_segment_size=min_segment_size,
min_absolute_change=min_absolute_change,
min_relative_change=min_relative_change,
min_steppiness=min_steppiness,
multiple_of_std_dev=multiple_of_std_dev)
actual_indexes = [a.x_value for a in results]
self.assertEqual(expected_indexes, actual_indexes)
def testFindChangePoints_EmptySeries(self):
# For an empty series, there are certainly no change points.
self.assertEqual([], find_change_points.FindChangePoints([]))
