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
|
import os
import shutil
import numpy as np
import pandas as pd
import torch
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import train_test_split
from examples.sentence_level.wmt_2018.common.util.draw import draw_scatterplot, print_stat
from examples.sentence_level.wmt_2018.common.util.normalizer import fit, un_fit
from examples.sentence_level.wmt_2018.common.util.postprocess import format_submission
from examples.sentence_level.wmt_2018.common.util.reader import read_annotated_file, read_test_file
from examples.sentence_level.wmt_2018.multilingual.monotransquest_config import TEMP_DIRECTORY, GOOGLE_DRIVE, DRIVE_FILE_ID, MODEL_NAME, \
monotransquest_config, MODEL_TYPE, SEED, RESULT_FILE, SUBMISSION_FILE
from transquest.algo.sentence_level.monotransquest.evaluation import pearson_corr, spearman_corr
from transquest.algo.sentence_level.monotransquest.run_model import MonoTransQuestModel
if not os.path.exists(TEMP_DIRECTORY):
os.makedirs(TEMP_DIRECTORY)
languages = {
# "DE-EN": ["examples/sentence_level/wmt_2018/de_en/data/de_en/",
# "examples/sentence_level/wmt_2018/de_en/data/de_en/",
# "examples/sentence_level/wmt_2018/de_en/data/de_en/",
# "smt"],
"EN-ZH": ["examples/sentence_level/wmt_2020_task2/en_zh/data/en-zh/train",
"examples/sentence_level/wmt_2020_task2/en_zh/data/en-zh/dev",
"examples/sentence_level/wmt_2020_task2/en_zh/data/en-zh/test-blind",
""],
"EN-CS": ["examples/sentence_level/wmt_2018/en_cs/data/en_cs/",
"examples/sentence_level/wmt_2018/en_cs/data/en_cs/",
"examples/sentence_level/wmt_2018/en_cs/data/en_cs/",
"smt"],
"EN-DE-NMT": ["examples/sentence_level/wmt_2020_task2/en_de/data/en-de/train",
"examples/sentence_level/wmt_2020_task2/en_de/data/en-de/dev",
"examples/sentence_level/wmt_2020_task2/en_de/data/en-de/test-blind",
""],
"EN-DE-SMT": ["examples/sentence_level/wmt_2018/en_de/data/en_de",
"examples/sentence_level/wmt_2018/en_de/data/en_de",
"examples/sentence_level/wmt_2018/en_de/data/en_de",
"smt"],
"EN-RU": ["examples/sentence_level/wmt_2019/en_ru/data/en-ru/train",
"examples/sentence_level/wmt_2019/en_ru/data/en-ru/dev",
"examples/sentence_level/wmt_2019/en_ru/data/en-ru/test-blind",
""],
"EN-LV-NMT": ["examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"nmt"],
"EN-LV-SMT": ["examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"examples/sentence_level/wmt_2018/en_lv/data/en_lv",
"smt"],
}
train_list = []
dev_list = []
test_list = []
index_list = []
test_sentence_pairs_list = []
for key, value in languages.items():
if value[3] == "nmt":
train_temp = read_annotated_file(path=value[0], original_file="train.nmt.src", translation_file="train.nmt.mt",
hter_file="train.nmt.hter")
dev_temp = read_annotated_file(path=value[1], original_file="dev.nmt.src", translation_file="dev.nmt.mt",
hter_file="dev.nmt.hter")
test_temp = read_test_file(path=value[2], original_file="test.nmt.src", translation_file="test.nmt.mt")
elif value[3] == "smt":
train_temp = read_annotated_file(path=value[0], original_file="train.smt.src", translation_file="train.smt.mt",
hter_file="train.smt.hter")
dev_temp = read_annotated_file(path=value[1], original_file="dev.smt.src", translation_file="dev.smt.mt",
hter_file="dev.smt.hter")
test_temp = read_test_file(path=value[2], original_file="test.smt.src", translation_file="test.smt.mt")
else:
train_temp = read_annotated_file(path=value[0], original_file="train.src", translation_file="train.mt",
hter_file="train.hter")
dev_temp = read_annotated_file(path=value[1], original_file="dev.src", translation_file="dev.mt",
hter_file="dev.hter")
test_temp = read_test_file(path=value[2], original_file="test.src", translation_file="test.mt")
train_temp = train_temp[['original', 'translation', 'hter']]
dev_temp = dev_temp[['original', 'translation', 'hter']]
test_temp = test_temp[['index', 'original', 'translation']]
index_temp = test_temp['index'].to_list()
train_temp = train_temp.rename(columns={'original': 'text_a', 'translation': 'text_b', 'hter': 'labels'}).dropna()
dev_temp = dev_temp.rename(columns={'original': 'text_a', 'translation': 'text_b', 'hter': 'labels'}).dropna()
test_temp = test_temp.rename(columns={'original': 'text_a', 'translation': 'text_b'}).dropna()
test_sentence_pairs_temp = list(map(list, zip(test_temp['text_a'].to_list(), test_temp['text_b'].to_list())))
train_temp = fit(train_temp, 'labels')
dev_temp = fit(dev_temp, 'labels')
train_list.append(train_temp)
dev_list.append(dev_temp)
test_list.append(test_temp)
index_list.append(index_temp)
test_sentence_pairs_list.append(test_sentence_pairs_temp)
train = pd.concat(train_list)
if monotransquest_config["evaluate_during_training"]:
if monotransquest_config["n_fold"] > 1:
dev_preds_list = []
test_preds_list = []
for dev, test in zip(dev_list, test_list):
dev_preds = np.zeros((len(dev), monotransquest_config["n_fold"]))
test_preds = np.zeros((len(test), monotransquest_config["n_fold"]))
dev_preds_list.append(dev_preds)
test_preds_list.append(test_preds)
for i in range(monotransquest_config["n_fold"]):
if os.path.exists(monotransquest_config['output_dir']) and os.path.isdir(monotransquest_config['output_dir']):
shutil.rmtree(monotransquest_config['output_dir'])
model = MonoTransQuestModel(MODEL_TYPE, MODEL_NAME, num_labels=1, use_cuda=torch.cuda.is_available(),
args=monotransquest_config)
train_df, eval_df = train_test_split(train, test_size=0.1, random_state=SEED * i)
model.train_model(train_df, eval_df=eval_df, pearson_corr=pearson_corr, spearman_corr=spearman_corr,
mae=mean_absolute_error)
model = MonoTransQuestModel(MODEL_TYPE, monotransquest_config["best_model_dir"], num_labels=1,
use_cuda=torch.cuda.is_available(), args=monotransquest_config)
for dev, test_sentence_pairs, dev_preds, test_preds in zip(dev_list, test_sentence_pairs_list, dev_preds_list, test_preds_list):
result, model_outputs, wrong_predictions = model.eval_model(dev, pearson_corr=pearson_corr,
spearman_corr=spearman_corr,
mae=mean_absolute_error)
predictions, raw_outputs = model.predict(test_sentence_pairs)
dev_preds[:, i] = model_outputs
test_preds[:, i] = predictions
for dev, dev_preds, test, test_preds in zip(dev_list, dev_preds_list, test_list, test_preds_list):
dev['predictions'] = dev_preds.mean(axis=1)
test['predictions'] = test_preds.mean(axis=1)
else:
model = MonoTransQuestModel(MODEL_TYPE, MODEL_NAME, num_labels=1, use_cuda=torch.cuda.is_available(),
args=monotransquest_config)
train_df, eval_df = train_test_split(train, test_size=0.1, random_state=SEED)
model.train_model(train_df, eval_df=eval_df, pearson_corr=pearson_corr, spearman_corr=spearman_corr,
mae=mean_absolute_error)
model = MonoTransQuestModel(MODEL_TYPE, monotransquest_config["best_model_dir"], num_labels=1,
use_cuda=torch.cuda.is_available(), args=monotransquest_config)
for dev, test, test_sentence_pairs in zip(dev_list, test_list, test_sentence_pairs_list):
result, model_outputs, wrong_predictions = model.eval_model(dev, pearson_corr=pearson_corr,
spearman_corr=spearman_corr,
mae=mean_absolute_error)
predictions, raw_outputs = model.predict(test_sentence_pairs)
dev['predictions'] = model_outputs
test['predictions'] = predictions
else:
model = MonoTransQuestModel(MODEL_TYPE, MODEL_NAME, num_labels=1, use_cuda=torch.cuda.is_available(),
args=monotransquest_config)
model.train_model(train, pearson_corr=pearson_corr, spearman_corr=spearman_corr, mae=mean_absolute_error)
for dev, test, test_sentence_pairs in zip(dev_list, test_list, test_sentence_pairs_list):
result, model_outputs, wrong_predictions = model.eval_model(dev, pearson_corr=pearson_corr,
spearman_corr=spearman_corr,
mae=mean_absolute_error)
predictions, raw_outputs = model.predict(test_sentence_pairs)
dev['predictions'] = model_outputs
test['predictions'] = predictions
for dev, test, index, language in zip(dev_list, test_list, index_list, [*languages]):
dev = un_fit(dev, 'labels')
dev = un_fit(dev, 'predictions')
test = un_fit(test, 'predictions')
dev.to_csv(os.path.join(TEMP_DIRECTORY, RESULT_FILE.split(".")[0] + "_" + language + "." + RESULT_FILE.split(".")[1]), header=True, sep='\t', index=False, encoding='utf-8')
print(language)
print_stat(dev, 'labels', 'predictions')
format_submission(df=test, index=index, method="TransQuest",
path=os.path.join(TEMP_DIRECTORY, SUBMISSION_FILE.split(".")[0] + "_" + language + "." +
SUBMISSION_FILE.split(".")[1]))
|
