-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmt_utils.py
339 lines (275 loc) · 11.3 KB
/
mt_utils.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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
import glob
from io import StringIO
import pandas as pd
import os
from scipy.stats import pearsonr
def find_corpus(name):
WMT2017 = dict({
"newstest2017-csen-ref.en": "cs-en",
"newstest2017-deen-ref.en": "de-en",
"newstest2017-fien-ref.en": "fi-en",
"newstest2017-lven-ref.en": "lv-en",
"newstest2017-ruen-ref.en": "ru-en",
"newstest2017-tren-ref.en": "tr-en",
"newstest2017-zhen-ref.en": "zh-en"
})
WMT2018 = dict({
"newstest2018-csen-ref.en": "cs-en",
"newstest2018-deen-ref.en": "de-en",
"newstest2018-eten-ref.en": "et-en",
"newstest2018-fien-ref.en": "fi-en",
"newstest2018-ruen-ref.en": "ru-en",
"newstest2018-tren-ref.en": "tr-en",
"newstest2018-zhen-ref.en": "zh-en",
})
WMT2019 = dict({
"newstest2019-deen-ref.en": "de-en",
"newstest2019-fien-ref.en": "fi-en",
"newstest2019-guen-ref.en": "gu-en",
"newstest2019-kken-ref.en": "kk-en",
"newstest2019-lten-ref.en": "lt-en",
"newstest2019-ruen-ref.en": "ru-en",
"newstest2019-zhen-ref.en": "zh-en",
})
if name == 'WMT17':
dataset = WMT2017
if name == 'WMT18':
dataset = WMT2018
if name == 'WMT19':
dataset = WMT2019
return dataset
def load_data(path):
lines = []
with open(path, 'r', encoding='utf8') as f:
for line in f.readlines():
l = line.strip()
lines.append(l)
return lines
def load_metadata(lp):
files_path = []
for root, directories, files in os.walk(lp):
for file in files:
if '.hybrid' not in file:
raw = file.split('.')
testset = raw[0]
lp = raw[-1]
system = '.'.join(raw[1:-1])
files_path.append((os.path.join(root, file), testset, lp, system))
return files_path
def df_append(metric, num_samples, lp, testset, system, score):
return pd.DataFrame({'metric': [metric] * num_samples,
'lp': [lp] * num_samples,
'testset': [testset] * num_samples,
'system': [system] * num_samples,
'sid': [_ for _ in range(1, num_samples + 1)],
'score': score,
})
def pearson_and_spearman(preds, labels):
pearson_corr = pearsonr(preds, labels)[0]
return str('{0:.{1}f}'.format(pearson_corr, 3))
def output_MT_sys_level_correlation(lp_set, eval_metric, f):
submissions = ["%s.sys.score" % (eval_metric)]
lines = [line.rstrip('\n') for line in open(f)]
lines.pop(0)
manual = {}
for l in lines:
l = l.replace("nmt-smt-hybrid", "nmt-smt-hybr")
c = l.split()
if len(c) != 3:
print("erorr in manual evaluation file")
exit(1)
lp = c[0]
score = c[1]
system = c[2]
if lp not in manual:
manual[lp] = {}
if system not in manual[lp]:
manual[lp][system] = score
missing = 0
met_names = {}
lms = {}
lsm = {}
for s in submissions:
files = glob.glob(s)
for f in files:
lines = [line.rstrip('\n') for line in open(f)]
for l in lines:
l = l.replace("nmt-smt-hybrid", "nmt-smt-hybr")
if (l.find("hybrid") == -1) and (l.find("himl") == -1):
c = l.split()
if ((len(c) != 5) and len(c) != 7) and (len(c) != 9):
missing = missing + 1
else:
metric, lp, data, system, score = c[0], c[1], c[2], c[3], c[4]
if data not in ["newstest2017", "newstest2018"]:
print("error with data set for metric: " + l)
exit(1)
if lp not in lms:
lms[lp] = {}
if metric not in lms[lp]:
lms[lp][metric] = {}
if system not in lms[lp][metric]:
lms[lp][metric][system] = score
if lp not in lsm:
lsm[lp] = {}
if system not in lsm[lp]:
lsm[lp][system] = {}
if system not in lsm[lp][system]:
lsm[lp][system][metric] = score
for lp in manual:
if lp not in lp_set: continue
for metric in lms[lp]:
if sorted(lms[lp][metric]) == sorted(manual[lp]):
if lp not in met_names:
met_names[lp] = {}
if metric not in met_names[lp]:
met_names[lp][metric] = 1
else:
print("systems mismatch " + lp + " " + metric)
print(sorted(lms[lp][metric]))
print(sorted(manual[lp]))
res_str = ""
for lp in manual:
if lp not in lp_set: continue
l = lp.replace("-", "")
s = "LP SYSTEM HUMAN"
for metric in sorted(met_names[lp]):
s = s + " " + metric
s = s + "\n"
for system in manual[lp]:
s = s + lp + " " + system + " " + manual[lp][system]
for metric in sorted(met_names[lp]):
s = s + " " + lsm[lp][system][metric]
s = s + "\n"
results = pd.read_csv(StringIO(s), sep=" ")
res_str = res_str + eval_metric + '\t' + lp + "\t" + pearson_and_spearman(results['HUMAN'],
results[eval_metric]) + "\n"
return pd.read_csv(StringIO(res_str), sep="\t", header=None)
import gzip
def output_MT_seg_level_correlation(lp_set, eval_metric, f):
submissions = eval_metric
lines = [line.rstrip('\n') for line in open(f)]
lines.pop(0)
manual = {}
for l in lines:
l = l.replace("nmt-smt-hybrid", "nmt-smt-hybr")
c = l.split()
if len(c) != 5:
print("error in manual evaluation file")
exit(1)
lp = c[0]
data = c[1]
sid = c[2]
better = c[3]
worse = c[4]
if lp not in manual:
manual[lp] = {}
if sid not in manual[lp]:
manual[lp][sid] = {}
if better not in manual[lp][sid]:
manual[lp][sid][better] = {}
if worse not in manual[lp][sid][better]:
manual[lp][sid][better][worse] = 1
missing = 0
met_names = {}
metrics = {}
for s in submissions:
files = glob.glob(s)
for f in files:
lines = [str(line, encoding='utf-8') for line in gzip.open(f)]
for l in lines:
l = l.replace("nmt-smt-hybrid", "nmt-smt-hybr")
if (l.find("hybrid") == -1) and (l.find("himl") == -1):
c = l.split()
if len(c) < 6:
missing = missing + 1
else:
metric = c[0]
lp = c[1]
data = c[2]
system = c[3]
sid = c[4]
score = float(c[5])
if data != "newstest2018":
continue
if lp not in metrics:
metrics[lp] = {}
if metric not in metrics[lp]:
metrics[lp][metric] = {}
if sid not in metrics[lp][metric]:
metrics[lp][metric][sid] = {}
if system not in metrics[lp][metric][sid]:
metrics[lp][metric][sid][system] = score
for lp in manual:
if lp not in lp_set: continue
if lp not in metrics:
print(lp + " not in metrics")
exit(1)
for metric in metrics[lp]:
allthere = True
for sid in manual[lp]:
if not sid in metrics[lp][metric]:
allthere = False
print("A) Missing " + lp + " " + metric + " " + sid + " no scores at all for this metric and sid")
else:
for s1 in manual[lp][sid]:
if not s1 in metrics[lp][metric][sid]:
allthere = False
print(
"B) Missing " + lp + " " + metric + " " + sid + " " + s1 + " no scores for this metric for sid and first system")
for s2 in manual[lp][sid][s1]:
if not s2 in metrics[lp][metric][sid]:
allthere = False
print(
"C) Missing " + lp + " " + metric + " " + sid + " " + s1 + " " + s2 + " no scores for this metric for sid and second system")
if allthere:
if lp not in met_names:
met_names[lp] = {}
if metric not in met_names[lp]:
met_names[lp][metric] = 1
res_str = ""
for lp in manual:
if lp not in lp_set: continue
for metric in met_names[lp]:
conc = 0
disc = 0
for sid in manual[lp]:
s = s + lp + " " + sid + " "
for better in manual[lp][sid]:
for worse in manual[lp][sid][better]:
if better not in metrics[lp][metric][sid]:
print("error " + lp + " " + metric + " " + better)
score1 = metrics[lp][metric][sid][better]
score2 = metrics[lp][metric][sid][worse]
if score1 > score2:
conc = conc + 1
else:
disc = disc + 1
conc = float(conc)
disc = float(disc)
result = (conc - disc) / (conc + disc)
res_str = res_str + metric + '\t' + lp + "\t" + '{0:.{1}f}'.format(result, 3) + "\n"
return pd.read_csv(StringIO(res_str), sep="\t", header=None)
def print_sys_level_correlation(metric, data, lp_set, f="DA-syslevel.csv"):
results = pd.concat(data, ignore_index=True)
del results['sid']
results = results.groupby(['metric', 'lp', 'testset', 'system']).mean()
results = results.reset_index()
results.to_csv(metric + '.sys.score', sep='\t', index=False, header=False)
outputs = output_MT_sys_level_correlation(lp_set, metric, f)
s = ' '.join(lp_set) + '\n'
s = s + metric + ' ' + ' '.join(
[str('{0:.{1}f}'.format(outputs[(outputs[1] == lp)].values[0][-1], 3)) for lp in lp_set])
print(s)
import shutil
def print_seg_level_correlation(metric, data, lp_set, f="RR-seglevel.csv"):
results = pd.concat(data, ignore_index=True)
results.to_csv(metric + '.seg.score', sep='\t', index=False, header=False)
with open(metric + '.seg.score', 'rb') as f_in:
with gzip.open(metric + '.seg.score.gz', 'wb') as f_out:
shutil.copyfileobj(f_in, f_out)
outputs = output_MT_seg_level_correlation(lp_set, [metric + '.seg.score.gz'], f)
s = ' '.join(lp_set) + '\n'
s = s + metric + ' ' + ' '.join(
[str('{0:.{1}f}'.format(outputs[(outputs[1] == lp)].values[0][-1], 3)) for lp in lp_set])
print(s)