|
1
|
|
|
import abc |
|
2
|
|
|
|
|
3
|
|
|
|
|
4
|
|
|
class Dataset(abc.ABC): |
|
5
|
|
|
"""Base class for datasets. |
|
6
|
|
|
|
|
7
|
|
|
Attributes |
|
8
|
|
|
- `df` - :class:`pandas.DataFrame` that holds the actual data. |
|
9
|
|
|
|
|
10
|
|
|
- `target` - Column name of the variable to predict |
|
11
|
|
|
(ground truth) |
|
12
|
|
|
|
|
13
|
|
|
- `sensitive_attributes` - Column name of the |
|
14
|
|
|
sensitive attributes |
|
15
|
|
|
|
|
16
|
|
|
- `prediction` - Columns name of the |
|
17
|
|
|
prediction (optional) |
|
18
|
|
|
|
|
19
|
|
|
""" |
|
20
|
|
|
|
|
21
|
|
|
@abc.abstractmethod |
|
22
|
|
|
def __init__(self, target, sensitive_attributes, prediction=None): |
|
23
|
|
|
"""Load, preprocess and validate the dataset. |
|
24
|
|
|
|
|
25
|
|
|
:param target: Column name of the variable |
|
26
|
|
|
to predict (ground truth) |
|
27
|
|
|
:param sensitive_attributes: Column name of the |
|
28
|
|
|
sensitive attributes |
|
29
|
|
|
:param prediction: Columns name of the |
|
30
|
|
|
prediction (optional) |
|
31
|
|
|
:type target: str |
|
32
|
|
|
:type sensitive_attributes: list |
|
33
|
|
|
:type prediction: str |
|
34
|
|
|
""" |
|
35
|
|
|
|
|
36
|
|
|
self.df = self._load_data() |
|
37
|
|
|
|
|
38
|
|
|
self._preprocess() |
|
39
|
|
|
|
|
40
|
|
|
self._name = self.__doc__.splitlines()[0] |
|
41
|
|
|
|
|
42
|
|
|
self.target = target |
|
43
|
|
|
self.sensitive_attributes = sensitive_attributes |
|
44
|
|
|
self.prediction = prediction |
|
45
|
|
|
|
|
46
|
|
|
self._validate() |
|
47
|
|
|
|
|
48
|
|
|
def __str__(self): |
|
49
|
|
|
return ('<{} {} rows, {} columns' |
|
50
|
|
|
' in which {{{}}} are sensitive attributes>' |
|
51
|
|
|
.format(self._name, |
|
52
|
|
|
len(self.df), |
|
53
|
|
|
len(self.df.columns), |
|
54
|
|
|
', '.join(self.sensitive_attributes))) |
|
55
|
|
|
|
|
56
|
|
|
@abc.abstractmethod |
|
57
|
|
|
def _load_data(self): |
|
58
|
|
|
pass |
|
59
|
|
|
|
|
60
|
|
|
@abc.abstractmethod |
|
61
|
|
|
def _preprocess(self): |
|
62
|
|
|
pass |
|
63
|
|
|
|
|
64
|
|
|
@abc.abstractmethod |
|
65
|
|
|
def _validate(self): |
|
66
|
|
|
# pylint: disable=line-too-long |
|
67
|
|
|
|
|
68
|
|
|
assert self.target in self.df.columns,\ |
|
69
|
|
|
('the target label \'{}\' should be in the columns' |
|
70
|
|
|
.format(self.target)) |
|
71
|
|
|
|
|
72
|
|
|
assert all(attr in self.df.columns |
|
73
|
|
|
for attr in self.sensitive_attributes),\ |
|
74
|
|
|
('the sensitive attributes {{{}}} should be in the columns' |
|
75
|
|
|
.format(','.join(attr for attr in self.sensitive_attributes |
|
76
|
|
|
if attr not in self.df.columns))) |
|
77
|
|
|
|
|
78
|
|
|
# assert all(attr in SENSITIVE_ATTRIBUTES |
|
79
|
|
|
# for attr in self.sensitive_attributes),\ |
|
80
|
|
|
# ('the sensitive attributes {} can be only from {}.' # noqa |
|
81
|
|
|
# .format(self.sensitive_attributes, SENSITIVE_ATTRIBUTES)) |
|
82
|
|
|
|
ResponsiblyAI /
responsibly
