-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathseg_eval.py
199 lines (168 loc) · 11.3 KB
/
seg_eval.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
import pandas as pd
import os
from scipy.stats import kendalltau
def compute_seg_correlation(domain, metric, human_judgment):
"""
Compute segment-level Kendall's tau for
SacreBLEU, TER, CHRF2, BLEURT-20, COMET-MQM_2021, COMET-QE-MQM_2021.
Get the correlation for both newstest2021 and tedtalks datasets.
:param sting domain: domain to compute the correlation for ('newstest2021' or 'tedtalks')
:param sting metric: the metric to compute the correlation for
('sacre_BLEU', 'TER', 'CHRF2', 'BLEURT-20', 'COMET-MQM_2021', 'COMET-QE-MQM_2021')
:param sting human_judgment: human judgment type ('mqm', 'raw_da', 'z_da')
:return: None
"""
path = f'../Data/{domain}/{metric}'
data_dict = {}
if metric not in ['sacre_BLEU', 'TER', 'CHRF2']:
for file in os.listdir(path):
if file.endswith('.tsv'):
file_df = pd.read_csv(f'../Data/{domain}/{metric}/{file}', sep='\t', on_bad_lines='skip', keep_default_na=False)
if metric in ['BLEURT-20', 'COMET-MQM_2021']:
if human_judgment == 'mqm':
if domain == 'newstest2021':
ref_A = list(file_df[f'{metric}_ref_A'])
ref_B = list(file_df[f'{metric}_ref_B'])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
annotated_human_ratings, corresponding_ref_A, corresponding_ref_B = [], [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_ref_A.append(float(ref_A[id]))
corresponding_ref_B.append(float(ref_B[id]))
cor_ref_A, p_value_ref_A = kendalltau(corresponding_ref_A, annotated_human_ratings)
cor_ref_B, p_value_ref_B = kendalltau(corresponding_ref_B, annotated_human_ratings)
cor = (cor_ref_A + cor_ref_B) / 2
data_dict[file.split('_')[0]] = f'{cor:.3f}'
elif domain == 'tedtalks':
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(r'human_judgments_seg/all_TED_seg_mqm_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
cor, p_value = kendalltau(metric_scores, human_ratings)
data_dict[file.split('_')[0]] = f'{cor:.3f}'
else:
if domain == 'newstest2021' and 'metricsystem' not in file:
ref_A = list(file_df[f'{metric}_ref_A'])
ref_B = list(file_df[f'{metric}_ref_B'])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
annotated_human_ratings, corresponding_ref_A, corresponding_ref_B = [], [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_ref_A.append(float(ref_A[id]))
corresponding_ref_B.append(float(ref_B[id]))
cor_ref_A, p_value_ref_A = kendalltau(corresponding_ref_A, annotated_human_ratings)
cor_ref_B, p_value_ref_B = kendalltau(corresponding_ref_B, annotated_human_ratings)
cor = (cor_ref_A + cor_ref_B) / 2
data_dict[file.split('_')[0]] = f'{cor:.3f}'
if metric == 'COMET-QE-MQM_2021':
if human_judgment == 'mqm':
if domain == 'newstest2021':
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
annotated_human_ratings, corresponding_metric_scores = [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_metric_scores.append(float(metric_scores[id]))
cor, p_value = kendalltau(corresponding_metric_scores, annotated_human_ratings)
data_dict[file.split('_')[0]] = f'{cor:.3f}'
elif domain == 'tedtalks':
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(r'human_judgments_seg/all_TED_seg_mqm_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
cor, p_value = kendalltau(metric_scores, human_ratings)
data_dict[file.split('_')[0]] = f'{cor:.3f}'
else:
if domain == 'newstest2021' and 'metricsystem' not in file:
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file.split('_')[0]])
annotated_human_ratings, corresponding_metric_scores = [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_metric_scores.append(float(metric_scores[id]))
cor, p_value = kendalltau(corresponding_metric_scores, annotated_human_ratings)
data_dict[file.split('_')[0]] = f'{cor:.3f}'
else:
path = f'../Data/{domain}/traditional_metrics'
for file in os.listdir(path):
if file.endswith('.tsv'):
file_path = f'{path}/{file}'
file_df = pd.read_csv(file_path, sep='\t', on_bad_lines='skip', keep_default_na=False)
if human_judgment == 'mqm':
if domain == 'newstest2021':
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file[:-4]])
annotated_human_ratings, corresponding_metric_scores = [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_metric_scores.append(float(metric_scores[id]))
cor, p_value = kendalltau(corresponding_metric_scores, annotated_human_ratings)
data_dict[file[:-4]] = f'{cor:.3f}'
elif domain == 'tedtalks':
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(r'human_judgments_seg/all_TED_seg_mqm_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file[:-4]])
cor, p_value = kendalltau(metric_scores, human_ratings)
data_dict[file[:-4]] = f'{cor:.3f}'
else:
if domain == 'newstest2021' and 'metricsystem' not in file:
metric_scores = list(file_df[metric])
human_ratings_df = pd.read_csv(f'human_judgments_seg/all_news_seg_{human_judgment}_scores.tsv', sep='\t', on_bad_lines='skip', keep_default_na=False)
human_ratings = list(human_ratings_df[file[:-4]])
annotated_human_ratings, corresponding_metric_scores = [], []
for id, human_rating in enumerate(human_ratings):
if human_rating != 'None':
annotated_human_ratings.append(float(human_rating))
corresponding_metric_scores.append(float(metric_scores[id]))
cor, p_value = kendalltau(corresponding_metric_scores, annotated_human_ratings)
data_dict[file[:-4]] = f'{cor:.3f}'
print(data_dict)
all_values = []
for value in data_dict.values():
all_values.append(float(value))
if human_judgment == 'mqm':
average = sum(all_values) / 14
print(f'Average: {average:.3f}')
else:
average = sum(all_values) / 9
print(f'Average: {average:.3f}')
if __name__ == '__main__':
metrics = ['sacre_BLEU', 'TER', 'CHRF2', 'BLEURT-20', 'COMET-MQM_2021', 'COMET-QE-MQM_2021']
print("Segment-level Kendall's tau correlation with MQM scores")
print('newstest2021:')
print('==================')
for metric in metrics:
print(metric)
compute_seg_correlation('newstest2021', metric, 'mqm')
print('------------------')
print()
print('tedtalks:')
print('==================')
for metric in metrics:
print(metric)
compute_seg_correlation('tedtalks', metric, 'mqm')
print('------------------')
print()
print("Segment-level Kendall's tau correlation with raw DA scores")
print('newstest2021:')
print('==================')
for metric in metrics:
print(metric)
compute_seg_correlation('newstest2021', metric, 'raw_da')
print('------------------')
print()
print("Segment-level Kendall's tau correlation with z-normalized DA scores")
print('newstest2021:')
print('==================')
for metric in metrics:
print(metric)
compute_seg_correlation('newstest2021', metric, 'z_da')
print('------------------')