Iteratively increase sample size for good/bad classification.

scripts/slave/recipe_modules/auto_bisect/resources/significantly_different_test.py

+#!/usr/bin/env python
+"""Tests for significantly_different."""
+
+import os
+import sys
+import unittest
+
+# pylint: disable=relative-import
+import significantly_different
+
+
+class SignificantlyDifferentTest(unittest.TestCase):
+
+  def setUp(self):
+    self.conda_path = None
+    try:
+      import scipy  # pylint: disable=unused-variable
+      self.conda_path = sys.executable
+    except ImportError:
+      if os.path.exists(os.path.expanduser('~/conda-test/bin/python')):
+        self.conda_path = os.path.expanduser('~/conda-test/bin/python')
+
+  def test_basic_case(self):
+    try:
+      results = significantly_different.main(
+          ['', '[1, 2, 3, 3, 2, 1]', '[1, 2, 2, 2, 5, 0]', '0.05'],
+          self.conda_path)
+    except significantly_different.ScipyNotInstalledError:
+      # This is meant to let presubmit pass on CQ bots :( because they don't
+      # have scipy either directly or thorugh anaconda.
+      return
+
+    self.assertAlmostEqual(
+        0.40073980338363635,
+        results['mann_p_value'])
+
+if __name__ == '__main__':
+  unittest.main()