1
|
|
|
__all__ = ['COMPASDataset'] |
2
|
|
|
|
3
|
|
|
import numpy as np |
4
|
|
|
import pandas as pd |
5
|
|
|
from pkg_resources import resource_filename |
6
|
|
|
|
7
|
|
|
from ethically.dataset.core import Dataset |
8
|
|
|
|
9
|
|
|
|
10
|
|
|
COMPAS_PATH = resource_filename(__name__, |
11
|
|
|
'compas-scores-two-years.csv') |
12
|
|
|
|
13
|
|
|
|
14
|
|
|
class COMPASDataset(Dataset): |
15
|
|
|
"""ProPublica Recidivism/COMPAS Dataset. |
16
|
|
|
|
17
|
|
|
See :class:`~ethically.dataset.Dataset` for a description of |
18
|
|
|
the arguments and attributes. |
19
|
|
|
|
20
|
|
|
References: |
21
|
|
|
https://github.com/propublica/compas-analysis |
22
|
|
|
|
23
|
|
|
""" |
24
|
|
|
|
25
|
|
|
def __init__(self): |
26
|
|
|
super().__init__(target='is_recid', |
27
|
|
|
sensitive_attributes=['race', 'sex'], |
28
|
|
|
prediction=['y_pred', 'score_factor', |
29
|
|
|
'score_text']) |
30
|
|
|
|
31
|
|
|
def _load_data(self): |
32
|
|
|
return pd.read_csv(COMPAS_PATH) |
33
|
|
|
|
34
|
|
|
def _preprocess(self): |
35
|
|
|
"""Perform the same preprocessing as the original analysis. |
36
|
|
|
|
37
|
|
|
https://github.com/propublica/compas-analysis/blob/master/Compas%20Analysis.ipynb |
38
|
|
|
""" |
39
|
|
|
|
40
|
|
|
self.df = self.df[(self.df['days_b_screening_arrest'] <= 30) |
41
|
|
|
& (self.df['days_b_screening_arrest'] >= -30) |
42
|
|
|
& (self.df['is_recid'] != -1) |
43
|
|
|
& (self.df['c_charge_degree'] != 'O') |
44
|
|
|
& (self.df['score_text'] != 'N/A')] |
45
|
|
|
|
46
|
|
|
self.df['c_jail_out'] = pd.to_datetime(self.df['c_jail_out']) |
47
|
|
|
self.df['c_jail_in'] = pd.to_datetime(self.df['c_jail_in']) |
48
|
|
|
self.df['length_of_stay'] = (self.df['c_jail_out'] |
49
|
|
|
- self.df['c_jail_in']) |
50
|
|
|
|
51
|
|
|
self.df['score_factor'] = np.where(self.df['score_text'] |
52
|
|
|
!= 'Low', |
53
|
|
|
'HighScore', 'LowScore') |
54
|
|
|
self.df['y_pred'] = (self.df['score_factor'] == 'HighScore') |
55
|
|
|
|
56
|
|
|
def _validate(self): |
57
|
|
|
# pylint: disable=line-too-long |
58
|
|
|
super()._validate() |
59
|
|
|
|
60
|
|
|
assert len(self.df) == 6172, 'the number of rows should be 6172,'\ |
61
|
|
|
' but it is {}.'.format(len(self.df)) |
62
|
|
|
assert len(self.df.columns) == 56, 'the number of columns should be 56,'\ |
63
|
|
|
' but it is {}.'.format(len(self.df.columns)) |
64
|
|
|
|