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
|
import csv
import logging
import math
import os
import shutil
import numpy as np
from sklearn.model_selection import train_test_split
from torch.utils.data import DataLoader
from examples.sentence_level.wmt_2018.common.util.download import download_from_google_drive
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.de_en.siamesetransquest_config import TEMP_DIRECTORY, GOOGLE_DRIVE, DRIVE_FILE_ID, \
MODEL_NAME, siamesetransquest_config, SEED, RESULT_FILE, SUBMISSION_FILE, RESULT_IMAGE
from transquest.algo.sentence_level.siamesetransquest import LoggingHandler, SentencesDataset, \
SiameseTransQuestModel
from transquest.algo.sentence_level.siamesetransquest import models, losses
from transquest.algo.sentence_level.siamesetransquest.evaluation import EmbeddingSimilarityEvaluator
from transquest.algo.sentence_level.siamesetransquest.readers import QEDataReader
logging.basicConfig(format='%(asctime)s - %(message)s',
datefmt='%Y-%m-%d %H:%M:%S',
level=logging.INFO,
handlers=[LoggingHandler()])
if not os.path.exists(TEMP_DIRECTORY):
os.makedirs(TEMP_DIRECTORY)
if GOOGLE_DRIVE:
download_from_google_drive(DRIVE_FILE_ID, MODEL_NAME)
TRAIN_FOLDER = "examples/sentence_level/wmt_2018/de_en/data/de_en/"
DEV_FOLDER = "examples/sentence_level/wmt_2018/de_en/data/de_en/"
TEST_FOLDER = "examples/sentence_level/wmt_2018/de_en/data/de_en/"
train = read_annotated_file(path=TRAIN_FOLDER, original_file="train.smt.src", translation_file="train.smt.mt", hter_file="train.smt.hter")
dev = read_annotated_file(path=DEV_FOLDER, original_file="dev.smt.src", translation_file="dev.smt.mt", hter_file="dev.smt.hter")
test = read_test_file(path=TEST_FOLDER, original_file="test.smt.src", translation_file="test.smt.mt")
index = test['index'].to_list()
train = train[['original', 'translation', 'hter']]
dev = dev[['original', 'translation', 'hter']]
test = test[['original', 'translation']]
train = train.rename(columns={'original': 'text_a', 'translation': 'text_b', 'hter': 'labels'}).dropna()
dev = dev.rename(columns={'original': 'text_a', 'translation': 'text_b', 'hter': 'labels'}).dropna()
test = test.rename(columns={'original': 'text_a', 'translation': 'text_b'}).dropna()
train = fit(train, 'labels')
dev = fit(dev, 'labels')
if siamesetransquest_config["evaluate_during_training"]:
if siamesetransquest_config["n_fold"] > 0:
dev_preds = np.zeros((len(dev), siamesetransquest_config["n_fold"]))
test_preds = np.zeros((len(test), siamesetransquest_config["n_fold"]))
for i in range(siamesetransquest_config["n_fold"]):
if os.path.exists(siamesetransquest_config['best_model_dir']) and os.path.isdir(
siamesetransquest_config['best_model_dir']):
shutil.rmtree(siamesetransquest_config['best_model_dir'])
if os.path.exists(siamesetransquest_config['cache_dir']) and os.path.isdir(
siamesetransquest_config['cache_dir']):
shutil.rmtree(siamesetransquest_config['cache_dir'])
os.makedirs(siamesetransquest_config['cache_dir'])
train_df, eval_df = train_test_split(train, test_size=0.1, random_state=SEED * i)
train_df.to_csv(os.path.join(siamesetransquest_config['cache_dir'], "train.tsv"), header=True, sep='\t',
index=False, quoting=csv.QUOTE_NONE)
eval_df.to_csv(os.path.join(siamesetransquest_config['cache_dir'], "eval_df.tsv"), header=True, sep='\t',
index=False, quoting=csv.QUOTE_NONE)
dev.to_csv(os.path.join(siamesetransquest_config['cache_dir'], "dev.tsv"), header=True, sep='\t',
index=False, quoting=csv.QUOTE_NONE)
test.to_csv(os.path.join(siamesetransquest_config['cache_dir'], "test.tsv"), header=True, sep='\t',
index=False, quoting=csv.QUOTE_NONE)
sts_reader = QEDataReader(siamesetransquest_config['cache_dir'], s1_col_idx=0, s2_col_idx=1,
score_col_idx=2,
normalize_scores=False, min_score=0, max_score=1, header=True)
word_embedding_model = models.Transformer(MODEL_NAME, max_seq_length=siamesetransquest_config[
'max_seq_length'])
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(),
pooling_mode_mean_tokens=True,
pooling_mode_cls_token=False,
pooling_mode_max_tokens=False)
model = SiameseTransQuestModel(modules=[word_embedding_model, pooling_model])
train_data = SentencesDataset(sts_reader.get_examples('train.tsv'), model)
train_dataloader = DataLoader(train_data, shuffle=True,
batch_size=siamesetransquest_config['train_batch_size'])
train_loss = losses.CosineSimilarityLoss(model=model)
eval_data = SentencesDataset(examples=sts_reader.get_examples('eval_df.tsv'), model=model)
eval_dataloader = DataLoader(eval_data, shuffle=False,
batch_size=siamesetransquest_config['train_batch_size'])
evaluator = EmbeddingSimilarityEvaluator(eval_dataloader)
warmup_steps = math.ceil(
len(train_data) * siamesetransquest_config["num_train_epochs"] / siamesetransquest_config[
'train_batch_size'] * 0.1)
model.fit(train_objectives=[(train_dataloader, train_loss)],
evaluator=evaluator,
epochs=siamesetransquest_config['num_train_epochs'],
evaluation_steps=100,
optimizer_params={'lr': siamesetransquest_config["learning_rate"],
'eps': siamesetransquest_config["adam_epsilon"],
'correct_bias': False},
warmup_steps=warmup_steps,
output_path=siamesetransquest_config['best_model_dir'])
model = SiameseTransQuestModel(siamesetransquest_config['best_model_dir'])
dev_data = SentencesDataset(examples=sts_reader.get_examples("dev.tsv"), model=model)
dev_dataloader = DataLoader(dev_data, shuffle=False, batch_size=8)
evaluator = EmbeddingSimilarityEvaluator(dev_dataloader)
model.evaluate(evaluator,
result_path=os.path.join(siamesetransquest_config['cache_dir'], "dev_result.txt"))
test_data = SentencesDataset(examples=sts_reader.get_examples("test.tsv", test_file=True), model=model)
test_dataloader = DataLoader(test_data, shuffle=False, batch_size=8)
evaluator = EmbeddingSimilarityEvaluator(test_dataloader)
model.evaluate(evaluator,
result_path=os.path.join(siamesetransquest_config['cache_dir'], "test_result.txt"),
verbose=False)
with open(os.path.join(siamesetransquest_config['cache_dir'], "dev_result.txt")) as f:
dev_preds[:, i] = list(map(float, f.read().splitlines()))
with open(os.path.join(siamesetransquest_config['cache_dir'], "test_result.txt")) as f:
test_preds[:, i] = list(map(float, f.read().splitlines()))
dev['predictions'] = dev_preds.mean(axis=1)
test['predictions'] = test_preds.mean(axis=1)
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), header=True, sep='\t', index=False, encoding='utf-8')
draw_scatterplot(dev, 'labels', 'predictions', os.path.join(TEMP_DIRECTORY, RESULT_IMAGE), "German-English")
print_stat(dev, 'labels', 'predictions')
format_submission(df=test, index=index, method="SiameseTransQuest",
path=os.path.join(TEMP_DIRECTORY, SUBMISSION_FILE))
|
