-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun_discrete.py
154 lines (126 loc) · 6.78 KB
/
run_discrete.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
import math
import json
import time
import torch
import argparse
import numpy as np
from tqdm import tqdm
from pathlib import Path
from mollifiers.utils import discrete_result_statistics
from mollifiers.dataset import compas_dataset, adult_dataset, dutchlarge_dataset, german_dataset, dutch_dataset, cross_validate_dataset
from mollifiers.booster import DiscreteBoostedDensityEstimator
from mollifiers.leverage import ExactLeverageSchedule, RelativeLeverageSchedule
from mollifiers.hypothesis import SKLearnDTHypothesisClass, SKLearnDTUncaliHypothesisClass
def main() -> None:
parser = argparse.ArgumentParser( description='Fair Mollifiers for Discrete Features Example' )
# Constants
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
# Boosting settings
parser.add_argument('--boosting-steps', type=int, default=50, metavar='T',
help='specify number of boosting steps (default: 50)')
parser.add_argument('--init-sr', type=float, default=1, metavar='R',
help='initial density representation rate (default: 1)')
parser.add_argument('--sr', type=float, default=0.9, metavar='R',
help='designated representation rate (default: 0.9)')
parser.add_argument('--leverage', type=str, default='exact', metavar='L',
help='specify leveraging function (default: "exact")')
parser.add_argument('--max-depth', type=int, default=8, metavar='D',
help='specify decision tree maximum depth (default: 8)')
parser.add_argument('--hypothesis', type=str, default='dt', metavar='H',
help='weak learner hypothesis class ["dt", "nn"] (default: "dt")')
parser.add_argument('--clip', type=float, default=math.log(2), metavar='H',
help='weak learner hypothesis class ["dt", "nn"] (default: "dt")')
parser.add_argument('--equal-rr', action='store_true',
help='set initial distribution to have equal RR (default: "False")')
parser.add_argument('--mix', type=float, default=0,
help='mixing parameter for the initial distribution (default: 0)')
# Data selecting
parser.add_argument('--dataset', type=str, default='compas', metavar='D',
help='selected dataset ["compas", "adult", "german", "dutch"] (default: "compas")')
parser.add_argument('--sattr', nargs='+', type=str, default=['sex'],
help='specify column name as a sensitive attribute (default: ["sex"])')
# Data splitting
parser.add_argument('--cvsplits', type=int, default=5,
help='number of cross validation splits (default: 5)')
# Evaluation
parser.add_argument('--kmeans-clusters', type=int, default=4, metavar='K',
help='number of kmeans cluster fitted for evaluation (default: 4)')
# Files
parser.add_argument('--save-folder', type=str, default='.',
help='save folder (default: ".")')
parser.add_argument('--save-file', type=str, default='mollifier_res.json',
help='save file (default: "mollifier_res.json")')
args = parser.parse_args()
print('\n#########')
print(args)
print( f'using random seed {args.seed}' )
np.random.seed( args.seed )
torch.manual_seed( args.seed )
save_path = Path(args.save_folder, args.save_file)
save_path.parent.mkdir(parents=True, exist_ok=True)
print(f'saving results to {save_path}')
hypothesis_path = Path(args.save_folder, 'hypothesis', args.save_file.rsplit('.', 1)[0])
hypothesis_path.mkdir(parents=True, exist_ok=True)
print(f'saving hypothsis to {hypothesis_path}')
# Get selected dataset
if args.dataset == 'compas':
dataset = compas_dataset(args.sattr)
elif args.dataset == 'adult':
dataset = adult_dataset(args.sattr)
elif args.dataset == 'german':
dataset = german_dataset(args.sattr)
elif args.dataset == 'dutch':
dataset = dutchlarge_dataset(args.sattr)
else:
raise ValueError(f'Unknown Dataset String: {args.dataset}')
print(f'hypothesis max depth: {args.max_depth}')
if args.hypothesis == 'dt':
hypothesis_class = SKLearnDTHypothesisClass(args.clip, dataset.info, max_depth=args.max_depth)
elif args.hypothesis == 'dtu':
hypothesis_class = SKLearnDTUncaliHypothesisClass(args.clip, dataset.info, max_depth=args.max_depth)
if args.leverage == 'exact':
leverage_func = ExactLeverageSchedule(C=args.clip)
elif args.leverage == 'relative':
leverage_func = RelativeLeverageSchedule(C=args.clip)
else:
raise ValueError(f'Unknown Leverage String: {args.leverage}')
# Set up cross validation loop
stats = []
dataset_cv = cross_validate_dataset(dataset, n_splits=args.cvsplits, random_state=args.seed)
for cv_i, (cur_train_dataset, cur_test_dataset) in tqdm(enumerate(dataset_cv), total=5, desc='fold prog', position=0):
tau = args.sr
booster = DiscreteBoostedDensityEstimator(cur_train_dataset.data,
cur_train_dataset.info, tau,
hypothesis_class,
leverage_func, args.init_sr,
equal_rr = args.equal_rr,
mix=args.mix)
booster.init()
boost_stats = [ discrete_result_statistics(booster, cur_train_dataset, cur_test_dataset, 0, args.kmeans_clusters, args.seed) ]
boost_time = 0
for boost_iter in tqdm(range(args.boosting_steps), desc=' boost iter', position=1, leave=False):
# Computing samples for stats logging
### Boosting Step ###
cur_start_time = time.time()
booster.step()
cur_end_time = time.time()
### Boosting Step ###
# Record stats
boost_stats.append(discrete_result_statistics(booster, cur_train_dataset,
cur_test_dataset,
boost_iter+1,
args.kmeans_clusters,
args.seed))
boost_time += cur_end_time - cur_start_time
stats.append({
'cv_iter': cv_i,
'boost_stats': boost_stats,
'time': boost_time,
})
# Save
with save_path.open(mode='w') as f:
json.dump(stats, f, indent=4, separators=(',', ':'))
print('#########\n')
if __name__ == '__main__':
main()