/
laundryML.py
234 lines (194 loc) · 11.7 KB
/
laundryML.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
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
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
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
160
161
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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
#!/usr/bin/env python3
# coding=utf-8
"""
Created on Wed Jun 20 16:02:46 2018
"""
import os
import pandas as pd
import numpy as np
from sklearn.externals import joblib
from scipy.stats import entropy
from fairness_eval import ConfusionMatrix, FairnessMetric
from collections import namedtuple
from utils import *
class LaundryML(namedtuple('LaundryML', 'data_prefix test_prefix exp_prefix res_folder k opt rho beta metric maj_pos min_pos sensitve_attr non_sensitve_attr decision_attr')):
def run(self):
# experiments results urls
models_enum_file = './%s/%s/%s/model_enum.txt' % (self.res_folder, self.data_prefix, self.exp_prefix)
objectiveFunction_results = './%s/%s/%s/objectiveFunction.dump' % (self.res_folder, self.data_prefix, self.exp_prefix)
models_files = './%s/%s/%s/%s_corels.mdl' % (self.res_folder, self.data_prefix, self.exp_prefix, self.data_prefix)
# files for train performances
unfairness_train_results = './%s/%s/%s/unfairness_train.dump' % (self.res_folder, self.data_prefix, self.exp_prefix)
accuracy_train_results = './%s/%s/%s/accuracy_train.dump' % (self.res_folder, self.data_prefix, self.exp_prefix)
# files for test performances
unfairness_test_results = './%s/%s/%s/unfairness_test.dump' % (self.res_folder, self.data_prefix, self.exp_prefix)
accuracy_test_results = './%s/%s/%s/accuracy_test.dump' % (self.res_folder, self.data_prefix, self.exp_prefix)
# generate input files
genData(self.data_prefix, self.exp_prefix, self.test_prefix, self.res_folder)
train_file = './%s/%s/%s/%s_train.txt' % (self.res_folder, self.data_prefix, self.exp_prefix, self.data_prefix)
test_file = './%s/%s/%s/%s_test.txt' % (self.res_folder, self.data_prefix, self.exp_prefix, self.data_prefix)
# call of corels' enumerator
test_corels(self.data_prefix, self.exp_prefix, self.test_prefix, self.res_folder, rho=self.rho, beta=self.beta, k=self.k, opt=self.opt, metric=self.metric, min_pos=self.min_pos, maj_pos=self.maj_pos, branch_depth=np.inf)
#Récupération modèle et exportation dans fichier .txt
adult_mdl = joblib.load(models_files)
#Récupération prédiction et acc sur train et test set
pred_train, acc_train = adult_mdl.predict(train_file)
pred_test, acc_test = adult_mdl.predict(test_file)
# saving objective function values
joblib.dump(adult_mdl.obj_, objectiveFunction_results, compress=9)
# saving models enumaration file
with open(models_enum_file,'w') as mdl:
for k, pred_list in enumerate(adult_mdl.pred_description_):
mdl.write('Model %i: accuracy train=%f; accuracy test=%f; obj:%f \n' % (k, acc_train[k], acc_test[k],adult_mdl.obj_[k]))
for i in range(len(pred_list)):
pred = pred_list[i][1:-1]
#pred=pred.replace('0_is_False', self.prediction + '=Yes')
#pred=pred.replace('0_is_True', self.prediction + '=No')
#pred=pred.replace('1_is_False', self.prediction + '=No')
#pred=pred.replace('1_is_True', self.prediction + '=Yes')
if (i == 0):
rule = adult_mdl.rule_description_[k][i]
rule = str(rule).replace('{', '').replace('}', '').split(',')
mdl.write('IF ')
for j, r in enumerate(rule):
mdl.write(r)
if j < len(rule) - 1:
mdl.write(' AND ')
mdl.write(' THEN ')
mdl.write(pred)
mdl.write('\n')
elif i == len(adult_mdl.pred_description_[k]) - 1:
mdl.write('ELSE ')
mdl.write(pred)
mdl.write('\n')
else:
rule = adult_mdl.rule_description_[k][i].replace('{', '').replace('}', '').split(',')
mdl.write('ELSE IF ')
for j, r in enumerate(rule):
mdl.write(r)
if j < len(rule) - 1:
mdl.write(' AND ')
mdl.write(' THEN ')
mdl.write(pred)
mdl.write('\n')
mdl.write('\n')
mdl.close()
# performance on training set
data_file = './data/%s/processed/_auditing_train.csv' % self.data_prefix
dataset = pd.read_csv(data_file)
thruth_file = './data/%s/processed/_scores_train.csv' % self.data_prefix
thruth_dataset = pd.read_csv(thruth_file)
unfairness_train = []
for idx, pred_list in enumerate(adult_mdl.pred_description_):
decision = pred_train[:,idx].astype(int)
dataset[self.decision_attr] = decision
cm = ConfusionMatrix(dataset[self.sensitve_attr], dataset[self.non_sensitve_attr], dataset[self.decision_attr], thruth_dataset[self.decision_attr])
cm_majority, cm_minority = cm.get_matrix()
fm = FairnessMetric(cm_majority, cm_minority)
if (self.metric == 1):
unfairness_train.append(fm.statistical_parity())
if (self.metric == 2):
unfairness_train.append(fm.predictive_parity())
if (self.metric == 3):
unfairness_train.append(fm.predictive_equality())
if (self.metric == 4):
unfairness_train.append(fm.equal_opportunity())
# savaing unfairness and accurracy values on training set
joblib.dump(unfairness_train, unfairness_train_results, compress=9)
joblib.dump(acc_train, accuracy_train_results, compress=9)
# performance on test set
data_file = './data/%s/processed/_auditing_test.csv' % self.data_prefix
dataset = pd.read_csv(data_file)
thruth_file = './data/%s/processed/_scores_test.csv' % self.data_prefix
thruth_dataset = pd.read_csv(thruth_file)
unfairness_test = []
for idx, pred_list in enumerate(adult_mdl.pred_description_):
decision = pred_test[:,idx].astype(int)
dataset[self.decision_attr] = decision
cm = ConfusionMatrix(dataset[self.sensitve_attr], dataset[self.non_sensitve_attr], dataset[self.decision_attr], thruth_dataset[self.decision_attr])
cm_majority, cm_minority = cm.get_matrix()
fm = FairnessMetric(cm_majority, cm_minority)
if (self.metric == 1):
unfairness_test.append(fm.statistical_parity())
if (self.metric == 2):
unfairness_test.append(fm.predictive_parity())
if (self.metric == 3):
unfairness_test.append(fm.predictive_equality())
if (self.metric == 4):
unfairness_test.append(fm.equal_opportunity())
# savaing unfairness and accurracy values on training set
joblib.dump(unfairness_test, unfairness_test_results, compress=9)
joblib.dump(acc_test, accuracy_test_results, compress=9)
"""
def test_adult_local(data_prefix,test_prefix,exp_prefix,rho, beta, k):
# experiments results urls
models_enum_file = './res_local/%s/%s/model_enum.txt' % (data_prefix,exp_prefix)
objectiveFunction_results = './res_local/%s/%s/objectiveFunction.dump' % (data_prefix,exp_prefix)
models_files = './res_local/%s/%s/%s_corels.mdl' % (data_prefix, exp_prefix, data_prefix)
# files for train performances
unfairness_train_results = './res_local/%s/%s/unfairness_train.dump' % (data_prefix,exp_prefix)
accuracy_train_results = './res_local/%s/%s/accuracy_train.dump' % (data_prefix,exp_prefix)
#Algo
genData_local(data_prefix, exp_prefix, test_prefix)
train_file = './res_local/%s/%s/%s_train.txt' % (data_prefix, exp_prefix, data_prefix)
test_corels_local(data_prefix, exp_prefix, test_prefix, rho, beta, k, opt='-c 2 -p 1', branch_depth=np.inf)
#Récupération modèle et exportation dans fichier .txt
adult_mdl = joblib.load(models_files)
#Récupération prédiction et acc sur train et test set
pred_train, acc = adult_mdl.predict(train_file)
acc_train = adult_mdl.predict_local(train_file)
# saving objective function values
joblib.dump(adult_mdl.obj_, objectiveFunction_results, compress=9)
# saving models enumaration file
with open(models_enum_file,'w') as mdl:
for k, pred_list in enumerate(adult_mdl.pred_description_):
mdl.write('Model %i: accuracy train=%f; accuracy test=%f; obj:%f \n' % (k, acc_train[k], acc_train[k],adult_mdl.obj_[k]))
for i in range(len(pred_list)):
pred = pred_list[i][1:-1]
pred=pred.replace('0_is_False','Income:>50K')
pred=pred.replace('0_is_True','Income:<=50K')
if (i == 0):
#if np.isnan(adult_mdl.rule_description_[k][i]):
#continue
#raw_rule = adult_mdl.rule_description_[k][i]
#print(">>>>>>>>>>>"*5, raw_rule)
rule = adult_mdl.rule_description_[k][i]
rule = str(rule).replace('{', '').replace('}', '').split(',')
mdl.write('IF ')
for j, r in enumerate(rule):
mdl.write(r)
if j < len(rule) - 1:
mdl.write(' AND ')
mdl.write(' THEN ')
mdl.write(pred)
mdl.write('\n')
elif i == len(adult_mdl.pred_description_[k]) - 1:
mdl.write('ELSE ')
mdl.write(pred)
mdl.write('\n')
else:
rule = adult_mdl.rule_description_[k][i]
rule = str(rule).replace('{', '').replace('}', '').split(',')
mdl.write('ELSE IF ')
for j, r in enumerate(rule):
mdl.write(r)
if j < len(rule) - 1:
mdl.write(' AND ')
mdl.write(' THEN ')
mdl.write(pred)
mdl.write('\n')
mdl.write('\n')
mdl.close()
# performance on training set
data_file = './algs/corels/data/%s/local/_auditing_train.csv' % data_prefix
dataset = pd.read_csv(data_file)
unfairness_train = []
for idx, pred_list in enumerate(adult_mdl.pred_description_):
decision = pred_train[:,idx].astype(int)
dataset["score"] = decision
ff = FairnessEvaluator(dataset["gender:Female"], dataset["gender:Male"], dataset["score"])
unfairness_train.append(ff.demographic_parity_discrimination())
# savaing unfairness and accurracy values on training set
joblib.dump(unfairness_train, unfairness_train_results, compress=9)
joblib.dump(acc_train, accuracy_train_results, compress=9)
"""