ruby/tensorflow_privacy
1
0
Fork 0
forked from 626_privacy/tensorflow_privacy
Code Pull requests Activity

Internal change.

Browse source 
PiperOrigin-RevId: 325423652
This commit is contained in:
A. Unique TensorFlower 2020-08-07 06:11:28 -07:00
parent 5ad8676d38
commit 40419b56a3
1 changed files with 17 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
tensorflow_privacy/privacy/membership_inference_attack/data_structures_test.py
View file

@ -17,6 +17,7 @@
import os
import tempfile
from absl.testing import absltest
from absl.testing import parameterized
import numpy as np
from tensorflow_privacy.privacy.membership_inference_attack.data_structures import AttackInputData
from tensorflow_privacy.privacy.membership_inference_attack.data_structures import AttackResults
@ -27,6 +28,20 @@ from tensorflow_privacy.privacy.membership_inference_attack.data_structures impo
from tensorflow_privacy.privacy.membership_inference_attack.data_structures import SlicingFeature
class SingleSliceSpecTest(parameterized.TestCase):
def testStrEntireDataset(self):
self.assertEqual(str(SingleSliceSpec()), 'Entire dataset')
@parameterized.parameters(
(SlicingFeature.CLASS, 2, 'CLASS=2'),
(SlicingFeature.PERCENTILE, (10, 20), 'Loss percentiles: 10-20'),
(SlicingFeature.CORRECTLY_CLASSIFIED, True, 'CORRECTLY_CLASSIFIED=True'),
)
def testStr(self, feature, value, expected_str):
self.assertEqual(str(SingleSliceSpec(feature, value)), expected_str)
class AttackInputDataTest(absltest.TestCase):
def test_get_loss(self):
@ -181,6 +196,7 @@ class AttackResultsTest(absltest.TestCase):
self.assertEqual(results.get_result_with_max_attacker_advantage(),
self.perfect_classifier_result)
@absltest.skip('Will be enabled in the next CL')
def test_summary_by_slices(self):
results = AttackResults(
[self.perfect_classifier_result, self.random_classifier_result])
@ -197,6 +213,7 @@ class AttackResultsTest(absltest.TestCase):
'Highest advantage on slice SingleSliceSpec(Entire dataset) achieved ' +
'by AttackType.THRESHOLD_ATTACK with an advantage of 0.0')
@absltest.skip('Will be enabled in the next CL')
def test_summary_without_slices(self):
results = AttackResults(
[self.perfect_classifier_result, self.random_classifier_result])