def testGermanCreditDataCreation(self):
protected, nonProtected = germanCreditData.create(
"test_dataset_creator/GermanCredit_sex.csv",
"DurationMonth",
"CreditAmount",
"score",
"sex",
protectedAttribute=["female"])
self.assertEqual(310, len(nonProtected), "should have 310 males")
self.assertEqual(
0, countProtected(nonProtected),
"the nonProtected array should not contain protected candidates")
self.assertAlmostEqual(
0.744449197,
nonProtected[0].originalQualification,
places=9,
msg="best non-protected candidate has wrong score")
self.assertEqual(690, len(protected), "should have 690 females")
self.assertEqual(
690, countProtected(protected),
"the protected ranking should contain only protected candidates")
self.assertAlmostEqual(0.790280217,
protected[0].originalQualification,
places=9,
msg="best protected candidate has wrong score")
def testDatasetCreationGender(self):
# the csv file contains 310 male persons and 690 female
protected, nonProtected = compasData.createGender("ProPublica_sex.csv", "sex", "Violence_rawscore", "Recidivism_rawscore",
"priors_count")
self.assertEqual(1328, len(nonProtected), "should have 1328 females")
self.assertEqual(0, countProtected(nonProtected), "the nonProtected array should not contain protected candidates")
# for COMPAS we take the negativ recidivism score
self.assertEqual(1 - 0.038379531, nonProtected[0].originalQualification, "best non-protected candidate has wrong score")
self.assertEqual(5561, len(protected), "should have 5561 males")
self.assertEqual(5561, countProtected(protected), "the protected ranking should contain only protected candidates")
# for COMPAS we take the negativ recidivism score
self.assertEqual(1 - 0, protected[0].originalQualification, "best protected candidate has wrong score")
def testDatasetCreationRace(self):
# the csv file contains 310 male persons and 690 female
protected, nonProtected = compasData.createRace("ProPublica_race.csv", "race", "Violence_rawscore", "Recidivism_rawscore",
"priors_count")
self.assertEqual(3361, len(nonProtected), "should have 3361 non-blacks")
self.assertEqual(0, countProtected(nonProtected), "the nonProtected array should not contain protected candidates")
# for COMPAS we take the negativ recidivism score
self.assertEqual(1 - 0, nonProtected[0].originalQualification, "best non-protected candidate has wrong score")
self.assertEqual(3528, len(protected), "should have 3528 blacks")
self.assertEqual(3528, countProtected(protected), "the protected ranking should contain only protected candidates")
# for COMPAS we take the negativ recidivism score
self.assertEqual(1 - 0.012793177, protected[0].originalQualification, "best protected candidate has wrong score")
def fair_representation_condition(self, ranking):
"""
checks if a given ranking with tau_p protected candidates fairly represents the protected group. A
minimal proportion of protected candidates is defined in advance.
Parameters:
----------
ranking : [Candidate]
the set to be checked for fair representation
Return:
------
True if the ranking fairly represents the protected group, False otherwise
"""
t = len(ranking)
numberProtected = countProtected(ranking)
if self.__candidatesNeeded[t - 1] > numberProtected:
# not enough protected candidates in my ranking
return False
else:
return True
def percentageOfProtected(ranking):
return countProtected(ranking) / len(ranking)