-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcopd_data_main.py
181 lines (170 loc) · 6.38 KB
/
copd_data_main.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
from datetime import datetime
from typing import Optional
import numpy as np
from causal_models.scm import SCM
from gmm.copd_data import (
LinearCombinedDataCrossFitNuisanceEstimator,
LinearLearnedNuisanceEstimator,
)
from gmm.observational_two_covariates import (
ObservationalTwoCovariatesGMMEqs,
ObservationalTwoCovariatesSingleMomentGMMEqs,
)
from sample_revealer import SampleRevealerSelectionBias
from strategies import (
ExploreThenCommitStrategy,
ExploreThenGreedyStrategy,
OracleStrategy,
StrategyRunResult,
)
from utils.print_utils import print_and_log
def execute_strategy_iteration(
true_scm_val: SCM,
true_scm_main: SCM,
strategy_name: str,
iteration_num: int,
horizon: int,
cost_per_source: list[float],
optimal_kappa: Optional[float] = None,
logging_path: Optional[str] = None,
store_nuisance_df: bool = False,
# keep this because parallel_utils expects this arg.
true_scm: Optional[SCM] = None,
) -> StrategyRunResult:
if logging_path:
print_and_log(
"%s: Start %s, iter: %d, horizon %d"
% (datetime.now(), strategy_name, iteration_num, horizon),
filepath=f"{logging_path}_{strategy_name}_{horizon}",
)
random_seed = 232281293 + iteration_num
np.random.seed(random_seed)
num_samples = horizon
df_val = true_scm_val.generate_data_samples(
num_samples=int(num_samples / max(cost_per_source) + 1)
)
df_main = true_scm_main.generate_data_samples(num_samples=int(num_samples))
sample_revealer = SampleRevealerSelectionBias(
budget=num_samples, df_obs=df_val, df_bias=df_main
)
nuisance_endpoints = np.arange(35, num_samples * 2, step=15)
if strategy_name == "oracle":
assert optimal_kappa is not None, "`optimal_kappa` must be set."
nuisance = LinearCombinedDataCrossFitNuisanceEstimator()
strategy = OracleStrategy(
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
optimal_kappa=optimal_kappa,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "single_source":
nuisance = LinearCombinedDataCrossFitNuisanceEstimator()
strategy = OracleStrategy(
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesSingleMomentGMMEqs(),
nuisance=nuisance,
optimal_kappa=1,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etc_0.1":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenCommitStrategy(
exploration=0.1,
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etc_0.2":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenCommitStrategy(
exploration=0.2,
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etc_0.4":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenCommitStrategy(
exploration=0.4,
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etg_0.1":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenGreedyStrategy(
batch_fractions=[0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 0.9],
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etg_0.2":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenGreedyStrategy(
batch_fractions=[0.2, 0.4, 0.6, 0.8],
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
elif strategy_name == "etg_0.4":
nuisance = LinearLearnedNuisanceEstimator(
horizon=num_samples, batch_endpoints=nuisance_endpoints
)
strategy = ExploreThenGreedyStrategy(
batch_fractions=[0.4, 0.5, 0.6, 0.7, 0.8],
sample_revealer=sample_revealer,
true_scm=true_scm_val,
gmm_eqs=ObservationalTwoCovariatesGMMEqs(),
nuisance=nuisance,
horizon=num_samples,
cost_per_source=cost_per_source,
store_nuisance_df=store_nuisance_df,
)
else:
raise ValueError(f"invalid strategy_name: {strategy_name}")
result = strategy.execute_run()
np.random.seed(None)
if logging_path:
print_and_log(
"%s: End %s, iter: %d, horizon %d"
% (datetime.now(), strategy_name, iteration_num, horizon),
filepath=f"{logging_path}_{strategy_name}_{horizon}",
)
return result