-
Notifications
You must be signed in to change notification settings - Fork 5
/
metalearners.py
251 lines (159 loc) · 9.88 KB
/
metalearners.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
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
import numpy as np
import pandas as pd
from scipy.stats import norm, beta
from sklearn.ensemble import RandomForestRegressor
from quantile_forest import RandomForestQuantileRegressor
from sklearn.linear_model import QuantileRegressor
from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingRegressor
from models.drlearner import *
def conformal_metalearner_experiment(df, metalearner="DR", quantile_regression=True, alpha=0.1, test_frac=0.1):
if len(df)==2:
train_data1, test_data = df
else:
train_data1, test_data = train_test_split(df, test_size=test_frac, random_state=42)
train_data, calib_data = train_test_split(train_data1, test_size=0.25, random_state=42)
#X_train = train_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_train = train_data.filter(like = 'X').values
T_train = train_data[['T']].values.reshape((-1,))
Y_train = train_data[['Y']].values.reshape((-1,))
ps_train = train_data[['ps']].values
#X_calib = calib_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_calib = calib_data.filter(like = 'X').values
T_calib = calib_data[['T']].values.reshape((-1,))
Y_calib = calib_data[['Y']].values.reshape((-1,))
ps_calib = calib_data[['ps']].values
ITEcalib = calib_data[['Y1']].values.reshape((-1,)) - calib_data[['Y0']].values.reshape((-1,))
#X_test = test_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_test = test_data.filter(like = 'X').values
T_test = test_data[['T']].values.reshape((-1,))
Y_test = test_data[['Y']].values.reshape((-1,))
ps_test = test_data[['ps']].values
model = conformalMetalearner(alpha=alpha, base_learner="GBM",
quantile_regression=quantile_regression,
metalearner=metalearner)
model.fit(X_train, T_train, Y_train, ps_train)
model.conformalize(alpha, X_calib, T_calib, Y_calib, ps_calib, oracle=ITEcalib)
T_hat_DR, T_hat_DR_l, T_hat_DR_u = model.predict(X_test)
True_effects = test_data[['Y1']].values.reshape((-1,)) - test_data[['Y0']].values.reshape((-1,))
CATE = test_data[['CATE']].values
conditional_coverage = np.mean((True_effects >= T_hat_DR_l) & (True_effects <= T_hat_DR_u))
average_interval_width = np.mean(np.abs(T_hat_DR_u - T_hat_DR_l))
PEHE = np.sqrt(np.mean((CATE-T_hat_DR)**2))
meta_conformity_score, oracle_conformity_score = model.residuals, model.oracle_residuals
conformity_scores = (meta_conformity_score, oracle_conformity_score)
return conditional_coverage, average_interval_width, PEHE, conformity_scores
def dr_cqr_random_forests(df, alpha):
if len(df)==2:
train_data1, test_data = df
else:
train_data1, test_data = train_test_split(df, test_size=test_frac, random_state=42)
train_data, calib_data = train_test_split(train_data1, test_size=0.25, random_state=42)
#X_train = train_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_train = train_data.filter(like = 'X').values
T_train = train_data[['T']].values.reshape((-1,))
Y_train = train_data[['Y']].values.reshape((-1,))
ps_train = train_data[['ps']].values
#X_calib = calib_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_calib = calib_data.filter(like = 'X').values
T_calib = calib_data[['T']].values.reshape((-1,))
Y_calib = calib_data[['Y']].values.reshape((-1,))
ps_calib = calib_data[['ps']].values
ITEcalib = calib_data[['Y1']].values.reshape((-1,)) - calib_data[['Y0']].values.reshape((-1,))
#X_test = test_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_test = test_data.filter(like = 'X').values
T_test = test_data[['T']].values.reshape((-1,))
Y_test = test_data[['Y']].values.reshape((-1,))
ps_test = test_data[['ps']].values
model = conformalMetalearner(alpha=alpha, base_learner="GBM",
quantile_regression=quantile_regression, metalearner="DR")
model.fit(X_train, T_train, Y_train, ps_train)
model.conformalize(alpha, X_calib, T_calib, Y_calib, ps_calib, oracle=ITEcalib)
T_hat_DR, T_hat_DR_l, T_hat_DR_u = model.predict(X_test)
True_effects = test_data[['Y1']].values.reshape((-1,)) - test_data[['Y0']].values.reshape((-1,))
CATE = test_data[['CATE']].values
conditional_coverage = np.mean((True_effects >= T_hat_DR_l) & (True_effects <= T_hat_DR_u))
average_interval_width = np.mean(np.abs(T_hat_DR_u - T_hat_DR_l))
PEHE = np.sqrt(np.mean((CATE-T_hat_DR)**2))
meta_conformity_score, oracle_conformity_score = model.residuals, model.oracle_residuals
conformity_scores = (meta_conformity_score, oracle_conformity_score)
return conditional_coverage, average_interval_width, PEHE, conformity_scores
def ipw_cqr_random_forests(df, alpha):
if len(df)==2:
train_data1, test_data = df
else:
train_data1, test_data = train_test_split(df, test_size=test_frac, random_state=42)
train_data, calib_data = train_test_split(train_data1, test_size=0.25, random_state=42)
#X_train = train_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_train = train_data.filter(like = 'X').values
T_train = train_data[['T']].values.reshape((-1,))
Y_train = train_data[['Y']].values.reshape((-1,))
ps_train = train_data[['ps']].values
#X_calib = calib_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_calib = calib_data.filter(like = 'X').values
T_calib = calib_data[['T']].values.reshape((-1,))
Y_calib = calib_data[['Y']].values.reshape((-1,))
ps_calib = calib_data[['ps']].values
ITEcalib = calib_data[['Y1']].values.reshape((-1,)) - calib_data[['Y0']].values.reshape((-1,))
#X_test = test_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_test = test_data.filter(like = 'X').values
T_test = test_data[['T']].values.reshape((-1,))
Y_test = test_data[['Y']].values.reshape((-1,))
ps_test = test_data[['ps']].values
model = conformalMetalearner(alpha=alpha, base_learner="GBM", quantile_regression=True, metalearner="IPW")
model.fit(X_train, T_train, Y_train, ps_train)
model.conformalize(alpha, X_calib, T_calib, Y_calib, ps_calib, oracle=ITEcalib)
T_hat_DR, T_hat_DR_l, T_hat_DR_u = model.predict(X_test)
True_effects = test_data[['Y1']].values.reshape((-1,)) - test_data[['Y0']].values.reshape((-1,))
CATE = test_data[['CATE']].values
conditional_coverage = np.mean((True_effects >= T_hat_DR_l) & (True_effects <= T_hat_DR_u))
average_interval_width = np.mean(np.abs(T_hat_DR_u - T_hat_DR_l))
PEHE = np.sqrt(np.mean((CATE-T_hat_DR)**2))
meta_conformity_score, oracle_conformity_score = model.residuals, model.oracle_residuals
conformity_scores = (meta_conformity_score, oracle_conformity_score)
return conditional_coverage, average_interval_width, PEHE, conformity_scores
def x_cqr_random_forests(df, alpha):
if len(df)==2:
train_data1, test_data = df
else:
train_data1, test_data = train_test_split(df, test_size=test_frac, random_state=42)
train_data, calib_data = train_test_split(train_data1, test_size=0.25, random_state=42)
#X_train = train_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_train = train_data.filter(like = 'X').values
T_train = train_data[['T']].values.reshape((-1,))
Y_train = train_data[['Y']].values.reshape((-1,))
ps_train = train_data[['ps']].values
#X_calib = calib_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_calib = calib_data.filter(like = 'X').values
T_calib = calib_data[['T']].values.reshape((-1,))
Y_calib = calib_data[['Y']].values.reshape((-1,))
ps_calib = calib_data[['ps']].values
ITEcalib = calib_data[['Y1']].values.reshape((-1,)) - calib_data[['Y0']].values.reshape((-1,))
#X_test = test_data[['X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10']].values
X_test = test_data.filter(like = 'X').values
T_test = test_data[['T']].values.reshape((-1,))
Y_test = test_data[['Y']].values.reshape((-1,))
ps_test = test_data[['ps']].values
model = conformalMetalearner(alpha=alpha, base_learner="GBM", quantile_regression=True, metalearner="X")
model.fit(X_train, T_train, Y_train, ps_train)
model.conformalize(alpha, X_calib, T_calib, Y_calib, ps_calib, oracle=ITEcalib)
T_hat_DR, T_hat_DR_l, T_hat_DR_u = model.predict(X_test)
True_effects = test_data[['Y1']].values.reshape((-1,)) - test_data[['Y0']].values.reshape((-1,))
CATE = test_data[['CATE']].values
conditional_coverage = np.mean((True_effects >= T_hat_DR_l) & (True_effects <= T_hat_DR_u))
average_interval_width = np.mean(np.abs(T_hat_DR_u - T_hat_DR_l))
PEHE = np.sqrt(np.mean((CATE-T_hat_DR)**2))
meta_conformity_score, oracle_conformity_score = model.residuals, model.oracle_residuals
conformity_scores = (meta_conformity_score, oracle_conformity_score)
return conditional_coverage, average_interval_width, PEHE, conformity_scores
def run(data, func, **kwargs): # alpha):
results = []
if type(data)==tuple:
for df_train, df_test in zip(data[0], data[1]):
result = func((df_train, df_test), **kwargs)# alpha)
results.append(result)
else:
for df in data:
result = func(df, **kwargs)# alpha)
results.append(result)
return results