-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiments_hm_test_ensembles.py
127 lines (119 loc) · 5.7 KB
/
experiments_hm_test_ensembles.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
from utils_data import create_calibrated_df, prepare_dataset_for_high_vs_middle_prediction
from utils import plot_calibration, check_calibration, evaluate_level_prediction_high_vs_middle
import numpy as np
from utils_constants import CORRECTNESS
random_state = 42
split = 'test'
# BERT
df = create_calibrated_df([
'output_bert_seed0_%s.csv' % split,
'output_bert_seed3_%s.csv' % split,
'output_bert_seed42_%s.csv' % split,
])
print('BERT: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/bert_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/bert_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# XLNet
df = create_calibrated_df([
'output_xlnet_seed_2_%s.csv' % split,
'output_xlnet_seed_3_%s.csv' % split,
'output_xlnet_seed_4_%s.csv' % split,
])
print('XLNet: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/xlnet_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/xlnet_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# DistilBERT
df = create_calibrated_df([
'output_distilbert_seed1_%s.csv' % split,
'output_distilbert_seed3_%s.csv' % split,
'output_distilbert_seed42_%s.csv' % split,
])
print('DistilBERT: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/distilbert_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/distilbert_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# BERT DistilBERT
df = create_calibrated_df([
'output_bert_seed0_%s.csv' % split,
'output_bert_seed3_%s.csv' % split,
'output_bert_seed42_%s.csv' % split,
'output_distilbert_seed1_%s.csv' % split,
'output_distilbert_seed3_%s.csv' % split,
'output_distilbert_seed42_%s.csv' % split,
])
print('BERT-DistilBERT: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/bert_distilbert_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/bert_distilbert_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# BERT XLNet
df = create_calibrated_df([
'output_bert_seed0_%s.csv' % split,
'output_bert_seed3_%s.csv' % split,
'output_bert_seed42_%s.csv' % split,
'output_xlnet_seed_2_%s.csv' % split,
'output_xlnet_seed_3_%s.csv' % split,
'output_xlnet_seed_4_%s.csv' % split,
])
print('BERT-XLNet: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/bert_xlnet_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/bert_xlnet_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# DistilBERT XLNet
df = create_calibrated_df([
'output_distilbert_seed1_%s.csv' % split,
'output_distilbert_seed3_%s.csv' % split,
'output_distilbert_seed42_%s.csv' % split,
'output_xlnet_seed_2_%s.csv' % split,
'output_xlnet_seed_3_%s.csv' % split,
'output_xlnet_seed_4_%s.csv' % split,
])
print('DistilBERT-XLNet: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/distilbert_xlnet_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/distilbert_xlnet_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()
# BERT - DistilBERT - XLNet
df = create_calibrated_df([
'output_bert_seed0_%s.csv' % split,
'output_bert_seed3_%s.csv' % split,
'output_bert_seed42_%s.csv' % split,
'output_distilbert_seed1_%s.csv' % split,
'output_distilbert_seed3_%s.csv' % split,
'output_distilbert_seed42_%s.csv' % split,
'output_xlnet_seed_2_%s.csv' % split,
'output_xlnet_seed_3_%s.csv' % split,
'output_xlnet_seed_4_%s.csv' % split,
])
print('BERT-DistilBERT-XLNet: QA TEST ACCURACY = %.4f' % float(np.mean(df[CORRECTNESS])))
output_filename = 'output/bert_distilbert_xlnet_ensemble_%s.txt' % split
output_file = open(output_filename, "w")
plot_calibration(df, 0.1, image_name='output_figures/bert_distilbert_xlnet_ensemble_%s.pdf' % split)
check_calibration(df, 0.1, output_file=output_file)
df = prepare_dataset_for_high_vs_middle_prediction(df, output_file=output_file, random_state=random_state)
evaluate_level_prediction_high_vs_middle(df, output_file=output_file)
output_file.close()