Skip to content

Collections

The collections module provides predefined collections of fairness scenarios, developed to maximize the diversity in how existing fair ML methods perform on them.

Prespecified Collections

fairml_datasets.collections.Corpus

Bases: Collection

The full corpus including all scenarios and datasets.

This collection contains all available datasets and their associated scenarios, providing a comprehensive set for fairness analysis across the entire corpus.

Source code in fairml_datasets/collections.py 
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
class Corpus(Collection):
    """
    The full corpus including all scenarios and datasets.

    This collection contains all available datasets and their associated scenarios,
    providing a comprehensive set for fairness analysis across the entire corpus.
    """

    def __init__(self, inclue_large_datasets=True):
        """
        Initialize the Corpus with all available datasets and scenarios.

        Args:
            inclue_large_datasets: Whether to include datasets marked as 'large'
        """
        all_scenarios = []
        all_datasets = Datasets(inclue_large_datasets=inclue_large_datasets)
        for dataset in all_datasets:
            all_scenarios += generate_dataset_scenarios(dataset=dataset)

        super().__init__(scenarios=all_scenarios)

Functions

__init__(inclue_large_datasets=True)

Initialize the Corpus with all available datasets and scenarios.

Parameters:

Name Type Description Default
inclue_large_datasets

Whether to include datasets marked as 'large'

 True
Source code in fairml_datasets/collections.py 
55
56
57
58
59
60
61
62
63
64
65
66
67
def __init__(self, inclue_large_datasets=True):
    """
    Initialize the Corpus with all available datasets and scenarios.

    Args:
        inclue_large_datasets: Whether to include datasets marked as 'large'
    """
    all_scenarios = []
    all_datasets = Datasets(inclue_large_datasets=inclue_large_datasets)
    for dataset in all_datasets:
        all_scenarios += generate_dataset_scenarios(dataset=dataset)

    super().__init__(scenarios=all_scenarios)

fairml_datasets.collections.DecorrelatedSmall

Bases: PrespecifiedCollection

Collection of De-Correlated Datasets with k = 5.

This corresponds to Scenarios described in Table 3 identified with a k.

Source code in fairml_datasets/collections.py 
70
71
72
73
74
75
76
77
78
79
80
81
82
83
class DecorrelatedSmall(PrespecifiedCollection):
    """
    Collection of De-Correlated Datasets with k = 5.

    This corresponds to Scenarios described in Table 3 identified with a k.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "hmda:applicant_sex_name;applicant_race_name_1",
        "stop_question_and_frisk_data:SUSPECT_SEX;SUSPECT_RACE_DESCRIPTION; SUSPECT_REPORTED_AGE",
        "folktables_acsemployment_small:RAC1P",
    ]

fairml_datasets.collections.DecorrelatedLarge

Bases: PrespecifiedCollection

Collection of De-Correlated Datasets with tau = 0.

This corresponds to Scenarios described in Table 3 identified with a tau.

Source code in fairml_datasets/collections.py 
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
class DecorrelatedLarge(PrespecifiedCollection):
    """
    Collection of De-Correlated Datasets with tau = 0.

    This corresponds to Scenarios described in Table 3 identified with a tau.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "hmda:applicant_sex_name;applicant_race_name_1",
        "stop_question_and_frisk_data:SUSPECT_SEX;SUSPECT_RACE_DESCRIPTION; SUSPECT_REPORTED_AGE",
        "folktables_acsemployment_small:RAC1P",
        "folktables_acstraveltime:RAC1P",
        "compas:sex;age",
        "folktables_acsincome_small:RAC1P",
        "compas_2_years:age",
        "communities_unnormalized:pct12-21",
        "arrhythmia:sex",
        "folktables_acspubliccoverage_small:RAC1P",
        "compas_2_years_violent:age",
        "south_german_credit:age;foreign_worker",
        "dutch:age",
        "folktables_acsmobility_small:RAC1P",
        "law_school_tensorflow:gender",
        "german_credit_onehot:<= 25 years",
        "communities:racePctAsian",
        "nursery:finance",
        "german_credit_numeric:age",
        "chicago_strategic_subject_list:RACE CODE CD",
    ]

fairml_datasets.collections.PermissivelyLicensedSmall

Bases: PrespecifiedCollection

Collection of Permissively Licensed Datasets with k = 5.

This corresponds to Scenarios described in Table 4 identified with a k.

Source code in fairml_datasets/collections.py 
119
120
121
122
123
124
125
126
127
128
129
130
131
132
class PermissivelyLicensedSmall(PrespecifiedCollection):
    """
    Collection of Permissively Licensed Datasets with k = 5.

    This corresponds to Scenarios described in Table 4 identified with a k.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "communities_unnormalized:pct12-21",
        "lipton_synthetic_hiring_dataset:sex",
        "bank:age;marital",
    ]

fairml_datasets.collections.PermissivelyLicensedLarge

Bases: PrespecifiedCollection

Collection of Permissively Licensed Datasets with tau = 0.

This corresponds to Scenarios described in Table 4 identified with a tau.

Source code in fairml_datasets/collections.py 
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
class PermissivelyLicensedLarge(PrespecifiedCollection):
    """
    Collection of Permissively Licensed Datasets with tau = 0.

    This corresponds to Scenarios described in Table 4 identified with a tau.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "communities_unnormalized:pct12-21",
        "lipton_synthetic_hiring_dataset:sex",
        "bank:age;marital",
        "german_credit_onehot:> 25 years",
        "folktables_acsincome:RAC1P",
        "south_german_credit:age",
        "folktables_acsemployment_small:RAC1P",
        "german_credit_numeric:age",
        "student:sex;age",
        "folktables_acstraveltime_small:RAC1P",
        "folktables_acspubliccoverage_small:RAC1P",
        "communities:agePct16t24",
        "folktables_acsmobility:RAC1P",
        "law_school_tensorflow:gender",
    ]

fairml_datasets.collections.PermissivelyLicensedFull

Bases: PrespecifiedCollection

Full collection of Permissively Licensed Datasets.

This corresponds to all Scenarios described in Table 4.

Source code in fairml_datasets/collections.py 
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
class PermissivelyLicensedFull(PrespecifiedCollection):
    """
    Full collection of Permissively Licensed Datasets.

    This corresponds to all Scenarios described in Table 4.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "communities_unnormalized:pct12-21",
        "lipton_synthetic_hiring_dataset:sex",
        "bank:age;marital",
        "german_credit_onehot:> 25 years",
        "folktables_acsincome:RAC1P",
        "south_german_credit:age",
        "folktables_acsemployment_small:RAC1P",
        "german_credit_numeric:age",
        "student:sex;age",
        "folktables_acstraveltime_small:RAC1P",
        "folktables_acspubliccoverage_small:RAC1P",
        "communities:agePct16t24",
        "folktables_acsmobility:RAC1P",
        "law_school_tensorflow:gender",
        "arrhythmia:sex",
        "adult:race",
        "nursery:finance;parents",
        "folktables_acsincome_small:RAC1P",
        "creditcard:SEX",
        "folktables_acsmobility_small:RAC1P",
        "student_language:age",
        "drug:ethnicity",
        "law_school_lequy:racetxt;male",
        "folktables_acstraveltime:RAC1P",
        "bank_additional_full:age;marital",
        "german_credit:foreign_worker",
        "generate_synthetic_data:s1",
        "bank_additional:age",
        "folktables_acsemployment:RAC1P",
        "bank_full:age",
    ]

fairml_datasets.collections.GeographicSmall

Bases: PrespecifiedCollection

Collection of Geographically Diverse Datasets with k = 5.

This corresponds to Scenarios described in Table 5 identified with a k.

Source code in fairml_datasets/collections.py 
205
206
207
208
209
210
211
212
213
214
215
216
217
218
class GeographicSmall(PrespecifiedCollection):
    """
    Collection of Geographically Diverse Datasets with k = 5.

    This corresponds to Scenarios described in Table 5 identified with a k.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "dutch:age;citizenship",
        "creditcard:SEX",
        "german_credit_onehot:> 25 years",
    ]

fairml_datasets.collections.GeographicLarge

Bases: PrespecifiedCollection

Collection of Geographically Diverse Datasets with tau = 0.

This corresponds to Scenarios described in Table 5 identified with a tau.

Source code in fairml_datasets/collections.py 
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
class GeographicLarge(PrespecifiedCollection):
    """
    Collection of Geographically Diverse Datasets with tau = 0.

    This corresponds to Scenarios described in Table 5 identified with a tau.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "dutch:age;citizenship",
        "creditcard:SEX",
        "german_credit_onehot:> 25 years",
        "student:sex",
    ]

fairml_datasets.collections.GeographicFull

Bases: PrespecifiedCollection

Full collection of Geographically Diverse Datasets.

This corresponds to all Scenarios described in Table 5.

Source code in fairml_datasets/collections.py 
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
class GeographicFull(PrespecifiedCollection):
    """
    Full collection of Geographically Diverse Datasets.

    This corresponds to all Scenarios described in Table 5.
    """

    scenario_ids = [
        "folktables_acspubliccoverage:RAC1P",
        "heart_disease:sex",
        "dutch:age;citizenship",
        "creditcard:SEX",
        "german_credit_onehot:> 25 years",
        "student:sex",
        "arrhythmia:sex",
        "nursery:finance;parents",
        "synth:sensible_feature",
        "drug:ethnicity",
    ]

Usage Examples

Using the Complete Corpus

from fairml_datasets.collections import Corpus

# Create the corpus (all available datasets and their scenarios)
corpus = Corpus(inclue_large_datasets=True)

# Iterate through all scenarios in the corpus
for scenario in corpus:
    print(f"Dataset: {scenario.dataset_id}")
    print(f"Sensitive columns: {scenario.sensitive_columns}")

    # Load the data
    df = scenario.load(stage="prepared")

Using Predefined Collections

from fairml_datasets.collections import DecorrelatedSmall, PermissivelyLicensedFull, GeographicLarge

# Use a small collection of decorrelated datasets
collection = DecorrelatedSmall()
print(f"Collection contains {len(collection)} scenarios")

# Or use the full collection of permissively licensed datasets
full_collection = PermissivelyLicensedFull()
print(f"Full collection contains {len(full_collection)} scenarios")

# Load and analyze datasets from the geographic collection
geo_collection = GeographicLarge()
for scenario in geo_collection:
    print(f"Dataset: {scenario.dataset_id}")
    print(f"Sensitive columns: {scenario.sensitive_columns}")

    # Load the data
    df = scenario.load(stage="prepared")