-
Notifications
You must be signed in to change notification settings - Fork 0
/
cost_exp1.py
141 lines (126 loc) · 6.82 KB
/
cost_exp1.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
'For Experiment 1 (the fixed-cost case) in the paper'
'The UCI datasets used here are from https://github.com/treforevans/uci_datasets'
from knn import train_KNN_RwR,test_KNN_RwR
from NN_method import test_loss_head_NN,train_triage,train_g,get_test_assignments_triage,get_test_assignments_full_no_wid,train_full_NN,train_head,test_logis_head,train_val_logis_head
from uci_datasets import Dataset
from selectivenet import train_Sel_NN_cost,test_Sel_NN,train_sel_head_cost
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
for data_name in ['concrete','wine','airfoil','energy','housing','solar','forest','parkinsons']:
data = Dataset(data_name)
full_NN_loss_L=[]
NN_triage_loss_L=[]
sel_NN_loss_L=[]
knn_loss_L=[]
full_sel_loss_L=[]
full_NN_coverage_L=[]
NN_triage_coverage_L=[]
sel_NN_coverage_L=[]
knn_coverage_L=[]
full_sel_loss_L=[]
full_sel_coverage_L=[]
full_logis_loss_L=[]
full_logis_coverage_L=[]
full_loss_loss_L=[]
full_loss_coverage_L=[]
constraint=1
cost_list=[0.2,0.5,1,2]
for cost in cost_list:
full_NN_loss=[]
full_NN_cal_loss=[]
NN_triage_loss=[]
sel_NN_loss=[]
full_sel_loss=[]
knn_loss=[]
sel_NN_coverage=[]
knn_coverage=[]
full_NN_coverage=[]
NN_triage_coverage=[]
full_sel_coverage=[]
full_logis_loss=[]
full_logis_coverage=[]
full_loss_loss=[]
full_loss_coverage=[]
for split in range(10):
X_train, Y_train, X_test, Y_test = data.get_split(split)
Y_train=np.squeeze(Y_train)
Y_test=np.squeeze(Y_test)
val=False
X_train_, X_val, Y_train_, Y_val = train_test_split(X_train, Y_train,test_size=2/9,random_state=42)
train_data={'X':X_train,'Y':Y_train}
train_part={'X':X_train_,'Y':Y_train_}
val_data={'X':X_val,'Y':Y_val}
test_data={'X':X_test,'Y':Y_test}
mnet_full=train_full_NN(train_data)
print('---')
print('NN+knnRej')
loss,coverage=get_test_assignments_full_no_wid(train_data,test_data,mnet_full,constraint,cost)
print('loss:',loss)
full_NN_loss.append(loss)
full_NN_coverage.append(coverage)
print('---')
print('Triage')
mnet=train_triage(train_data,constraint,val,cost)
gnet=train_g(train_data,constraint,mnet,val,False,cost)
loss,coverage=get_test_assignments_triage(test_data,mnet,gnet,constraint,cost)
NN_triage_loss.append(loss)
NN_triage_coverage.append(coverage)
print('---')
print('KNN')
knn,var_knn,ecdf=train_KNN_RwR(train_data)
loss,coverage=test_KNN_RwR(test_data,knn,var_knn,ecdf,constraint,cost)
knn_loss.append(loss)
knn_coverage.append(coverage)
print('---')
print('SelectiveNet')
sel_net=train_Sel_NN_cost(train_data,cost)
loss,coverage=test_Sel_NN(test_data,sel_net,cost)
sel_NN_loss.append(loss)
sel_NN_coverage.append(coverage)
print('---')
print('NN+SelRej')
mnet=train_full_NN(train_data)
sel_net=train_sel_head_cost(train_data,mnet,cost)
loss,coverage=test_Sel_NN(test_data,sel_net,cost,no_aux=True)
full_sel_loss.append(loss)
full_sel_coverage.append(coverage)
print('---')
print('NN+LogRej')
thrs_list=[0.5]
mnet,rej_net,thrs=train_val_logis_head(train_data,constraint,cost,thrs_list)
loss,coverage=test_logis_head(test_data,mnet,rej_net,cost,thrs)
full_logis_loss.append(loss)
full_logis_coverage.append(coverage)
print('---')
print('NN+LossRej')
mnet=train_full_NN(train_data)
loss_net,thrs=train_head(train_data,mnet,'loss',constraint,cost)
loss,coverage=test_loss_head_NN(test_data,mnet,loss_net,thrs,cost)
full_loss_loss.append(loss)
full_loss_coverage.append(coverage)
full_NN_loss_L.append("{:.2f}".format(np.mean(full_NN_loss))+'+/-'+"{:.2f}".format(np.std(full_NN_loss)))
NN_triage_loss_L.append("{:.2f}".format(np.mean(NN_triage_loss))+'+/-'+"{:.2f}".format(np.std(NN_triage_loss)))
sel_NN_loss_L.append("{:.2f}".format(np.mean(sel_NN_loss))+'+/-'+"{:.2f}".format(np.std(sel_NN_loss)))
knn_loss_L.append("{:.2f}".format(np.mean(knn_loss))+'+/-'+"{:.2f}".format(np.std(knn_loss)))
sel_NN_coverage_L.append("{:.2f}".format(np.mean(sel_NN_coverage))+'+/-'+"{:.2f}".format(np.std(sel_NN_coverage)))
knn_coverage_L.append("{:.2f}".format(np.mean(knn_coverage))+'+/-'+"{:.2f}".format(np.std(knn_coverage)))
full_NN_coverage_L.append("{:.2f}".format(np.mean(full_NN_coverage))+'+/-'+"{:.2f}".format(np.std(full_NN_coverage)))
NN_triage_coverage_L.append("{:.2f}".format(np.mean(NN_triage_coverage))+'+/-'+"{:.2f}".format(np.std(NN_triage_coverage)))
full_sel_loss_L.append("{:.2f}".format(np.mean(full_sel_loss))+'+/-'+"{:.2f}".format(np.std(full_sel_loss)))
full_logis_loss_L.append("{:.2f}".format(np.mean(full_logis_loss))+'+/-'+"{:.2f}".format(np.std(full_logis_loss)))
full_logis_coverage_L.append("{:.2f}".format(np.mean(full_logis_coverage))+'+/-'+"{:.2f}".format(np.std(full_logis_coverage)))
full_sel_coverage_L.append("{:.2f}".format(np.mean(full_sel_coverage))+'+/-'+"{:.2f}".format(np.std(full_sel_coverage)))
full_loss_loss_L.append("{:.2f}".format(np.mean(full_loss_loss))+'+/-'+"{:.2f}".format(np.std(full_loss_loss)))
full_loss_coverage_L.append("{:.2f}".format(np.mean(full_loss_coverage))+'+/-'+"{:.2f}".format(np.std(full_loss_coverage)))
print('_____________________________________________________________')
print('_____________________________________________________________')
print('_____________________________________________________________')
print('____________________Finishied_________________________')
df=pd.DataFrame({'NN+knnRej':full_NN_loss_L,'NN+SelRej':full_sel_loss_L,'NN+LossRej':full_loss_loss_L,
'NN+LogRej':full_logis_loss_L,'Triage':NN_triage_loss_L,
'SelNN':sel_NN_loss_L,'kNN':knn_loss_L,
'NN+knnRej':full_NN_coverage_L,'NN+SelRej':full_sel_coverage_L,'NN+LossRej':full_logis_coverage_L,
'NN+LogRej':full_logis_coverage_L,'Triage':NN_triage_coverage_L,
'SelNN':sel_NN_coverage_L,'kNN':knn_coverage_L, },index=cost_list)
df.to_csv(data_name+'_new.csv')