-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrun_robustness_semisynth_experiment.py
171 lines (156 loc) · 9.1 KB
/
run_robustness_semisynth_experiment.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
import os
import numpy as np
import sys
import pandas as pd
import torch
import data_processing.semisynth_dataloader_robust as ssdl
import data_processing.semisynth_subsets as sub
from baselines import run_U_learner, run_causal_forest_efficient, run_model_direct, run_tarnet
from run_baseline_models import run_model
n_iter = 10
num_agents = 40
baseline_type = sys.argv[1]
assert baseline_type in {'CausalForest', 'Direct', 'Iterative', 'TarNet'}
if baseline_type == 'TarNet':
os.environ["CUDA_VISIBLE_DEVICES"]="0"
torch.cuda.device(0)
assert(torch.cuda.is_available())
torch.manual_seed(42)
num_groups = int(sys.argv[2])
assert num_groups > 1
logits = [-1.5 + 3./(num_groups-1)*i for i in range(num_groups)]
subset = sys.argv[3]
assert subset in {'drug_possession', 'misdemeanor_under35'}
if subset == 'drug_possession':
SUB = sub.drug_possession
else:
SUB = sub.misdemeanor_under35
output_dir = 'semisynth_robust_' + subset + '_numgroups' + str(num_groups) + '/'
if not os.path.exists(output_dir):
os.makedirs(output_dir)
model_classes = ['LogisticRegression', 'DecisionTree', 'RandomForest']
metrics = dict()
if baseline_type == 'CausalForest':
ulearner_metrics = dict()
for model_class in model_classes:
metrics[model_class] = {'region_precisions': [], 'region_recalls': [], 'region_aucs': [], 'partition_accs': []}
if baseline_type == 'CausalForest':
ulearner_metrics[model_class] = {'region_precisions': [], 'region_recalls': [], 'region_aucs': [], 'partition_accs': []}
for i in range(n_iter):
dd = ssdl.load_semi_synthetic_compas_data(
seed=i,
num_agents=num_agents,
num_groups=num_groups,
logit_adjust=logits,
subset_func=SUB,
verbose=False, cache=False)
filename = baseline_type + '_iter' + str(i) + '_results.pkl'
if baseline_type == 'CausalForest':
ulearner_filename = 'ULearner_iter' + str(i) + '_results.pkl'
ulearner_results_dict = run_U_learner(
dd['X'], dd['d'], dd['t'],
dd['train_idxs'], dd['valid_idxs'], dd['test_idxs'],
output_dir + ulearner_filename,
dd['true_region_func'], dd['true_provider_split'],
beta=dd['true_beta'])
print('ULearner iter ' + str(i))
for model_class in model_classes:
print(model_class)
print('Region precision: ' + str(ulearner_results_dict[model_class]['region_precision']))
print('Region recall: ' + str(ulearner_results_dict[model_class]['region_recall']))
print('Region AUC: ' + str(ulearner_results_dict[model_class]['region_auc']))
print('Partition accuracy: ' + str(ulearner_results_dict[model_class]['partition_acc']))
ulearner_metrics[model_class]['region_precisions'].append(ulearner_results_dict[model_class]['region_precision'])
ulearner_metrics[model_class]['region_recalls'].append(ulearner_results_dict[model_class]['region_recall'])
ulearner_metrics[model_class]['region_aucs'].append(ulearner_results_dict[model_class]['region_auc'])
ulearner_metrics[model_class]['partition_accs'].append(ulearner_results_dict[model_class]['partition_acc'])
oracle_preds = ulearner_results_dict['all_resids_pred']
results_dict = run_causal_forest(
dd['X'], dd['d'], dd['t'],
dd['train_idxs'], dd['valid_idxs'], dd['test_idxs'],
oracle_preds,
output_dir + filename,
dd['true_region_func'], dd['true_provider_split'],
beta=dd['true_beta'])
elif baseline_type == 'Direct':
results_dict = run_model_direct(
dd['X'], dd['d'], dd['t'],
dd['train_idxs'], dd['valid_idxs'], dd['test_idxs'],
output_dir + filename,
dd['true_region_func'], dd['true_provider_split'],
beta=dd['true_beta'])
elif baseline_type == 'TarNet':
tarnet_dir = baseline_type + '_iter' + str(i) + '/'
if not os.path.exists(output_dir + tarnet_dir):
os.makedirs(output_dir + tarnet_dir)
results_dict = run_tarnet(dd['X'], dd['d'], dd['t'], dd['train_idxs'], dd['valid_idxs'], dd['test_idxs'],
output_dir + filename, output_dir + tarnet_dir, dd['true_region_func'],
dd['true_provider_split'], beta=dd['true_beta'])
else:
results_dict = dict()
for model_class in model_classes:
results_dict[model_class] = run_model(baseline_type, model_class,
dd['X'], dd['d'], dd['t'],
dd['train_idxs'], dd['valid_idxs'], dd['test_idxs'],
output_dir + filename,
dd['true_region_func'], dd['true_provider_split'],
beta=dd['true_beta'], n_iter=100,
outcome_model_class='LogisticRegression', verbose=False)
print(baseline_type + ' iter' + str(i))
for model_class in model_classes:
print(model_class)
print('Region precision: ' + str(results_dict[model_class]['region_precision']))
print('Region recall: ' + str(results_dict[model_class]['region_recall']))
print('Region AUC: ' + str(results_dict[model_class]['region_auc']))
metrics[model_class]['region_precisions'].append(results_dict[model_class]['region_precision'])
metrics[model_class]['region_recalls'].append(results_dict[model_class]['region_recall'])
metrics[model_class]['region_aucs'].append(results_dict[model_class]['region_auc'])
if results_dict[model_class]['partition_acc'] is not None:
print('Partition accuracy: ' + str(results_dict[model_class]['partition_acc']))
metrics[model_class]['partition_accs'].append(results_dict[model_class]['partition_acc'])
if baseline_type == 'CausalForest':
print('ULearner mean (std) across ' + str(n_iter) + ' iterations')
ulearner_df = None
for model_class in model_classes:
print(model_class)
print('Region precision: {0:.4f}'.format(np.mean(ulearner_metrics[model_class]['region_precisions'])) \
+ ' ({0:.4f})'.format(np.std(ulearner_metrics[model_class]['region_precisions'])))
print('Region recall: {0:.4f}'.format(np.mean(ulearner_metrics[model_class]['region_recalls'])) \
+ ' ({0:.4f})'.format(np.std(ulearner_metrics[model_class]['region_recalls'])))
print('Region AUC: {0:.4f}'.format(np.mean(ulearner_metrics[model_class]['region_aucs'])) \
+ ' ({0:.4f})'.format(np.std(ulearner_metrics[model_class]['region_aucs'])))
print('Partition accuracy: {0:.4f}'.format(np.mean(ulearner_metrics[model_class]['partition_accs'])) \
+ ' ({0:.4f})'.format(np.std(ulearner_metrics[model_class]['partition_accs'])))
class_df = pd.DataFrame({'Model class': [model_class for i in range(n_iter)], 'iter': np.arange(n_iter), \
'Region precision': ulearner_metrics[model_class]['region_precisions'], \
'Region recall': ulearner_metrics[model_class]['region_recalls'], \
'Region AUC': ulearner_metrics[model_class]['region_aucs'], \
'Partition accuracy': ulearner_metrics[model_class]['partition_accs']})
if ulearner_df is None:
ulearner_df = class_df
else:
ulearner_df = pd.concat((ulearner_df, class_df), ignore_index=True)
ulearner_df.to_csv(output_dir + 'ULearner_summary_metrics.csv', index=False)
print(baseline_type + ' mean (std) across ' + str(n_iter) + ' iterations')
baseline_df = None
for model_class in model_classes:
print(model_class)
print('Region precision: {0:.4f}'.format(np.mean(metrics[model_class]['region_precisions'])) \
+ ' ({0:.4f})'.format(np.std(metrics[model_class]['region_precisions'])))
print('Region recall: {0:.4f}'.format(np.mean(metrics[model_class]['region_recalls'])) \
+ ' ({0:.4f})'.format(np.std(metrics[model_class]['region_recalls'])))
print('Region AUC: {0:.4f}'.format(np.mean(metrics[model_class]['region_aucs'])) \
+ ' ({0:.4f})'.format(np.std(metrics[model_class]['region_aucs'])))
class_df = pd.DataFrame({'Model class': [model_class for i in range(n_iter)], 'iter': np.arange(n_iter), \
'Region precision': metrics[model_class]['region_precisions'], \
'Region recall': metrics[model_class]['region_recalls'], \
'Region AUC': metrics[model_class]['region_aucs']})
if len(metrics[model_class]['partition_accs']) > 0:
print('Partition accuracy: {0:.4f}'.format(np.mean(metrics[model_class]['partition_accs'])) \
+ ' ({0:.4f})'.format(np.std(metrics[model_class]['partition_accs'])))
class_df['Partition accuracy'] = metrics[model_class]['partition_accs']
if baseline_df is None:
baseline_df = class_df
else:
baseline_df = pd.concat((baseline_df, class_df), ignore_index=True)
baseline_df.to_csv(output_dir + baseline_type + '_summary_metrics.csv', index=False)