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
|
import os
import shutil
import numpy as np
import torch
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import train_test_split
from examples.sentence_level.wmt_2020.common.util.draw import draw_scatterplot, print_stat
from examples.sentence_level.wmt_2020.common.util.normalizer import fit, un_fit
from examples.sentence_level.wmt_2020.common.util.postprocess import format_submission
from examples.sentence_level.wmt_2020.common.util.reader import read_test_file, read_annotated_file
from examples.sentence_level.wmt_2020.ne_en.monotransquest_config import TEMP_DIRECTORY, MODEL_TYPE, MODEL_NAME, monotransquest_config, SEED, \
RESULT_FILE, RESULT_IMAGE, SUBMISSION_FILE, GOOGLE_DRIVE, DRIVE_FILE_ID
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)
TRAIN_FILE = "examples/sentence_level/wmt_2020/ne_en/data/ne-en/train.neen.df.short.tsv"
DEV_FILE = "examples/sentence_level/wmt_2020/ne_en/data/ne-en/dev.neen.df.short.tsv"
TEST_FILE = "examples/sentence_level/wmt_2020/ne_en/data/ne-en/test20.neen.df.short.tsv"
train = read_annotated_file(TRAIN_FILE)
dev = read_annotated_file(DEV_FILE)
test = read_test_file(TEST_FILE)
train = train[['original', 'translation', 'z_mean']]
dev = dev[['original', 'translation', 'z_mean']]
test = test[['index', 'original', 'translation']]
index = test['index'].to_list()
train = train.rename(columns={'original': 'text_a', 'translation': 'text_b', 'z_mean': 'labels'}).dropna()
dev = dev.rename(columns={'original': 'text_a', 'translation': 'text_b', 'z_mean': 'labels'}).dropna()
test = test.rename(columns={'original': 'text_a', 'translation': 'text_b'}).dropna()
test_sentence_pairs = list(map(list, zip(test['text_a'].to_list(), test['text_b'].to_list())))
train = fit(train, 'labels')
dev = fit(dev, 'labels')
assert (len(index) == 1000)
if monotransquest_config["evaluate_during_training"]:
if monotransquest_config["n_fold"] > 1:
dev_preds = np.zeros((len(dev), monotransquest_config["n_fold"]))
test_preds = np.zeros((len(test), monotransquest_config["n_fold"]))
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)
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
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)
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)
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
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), "Nepalese-English")
print_stat(dev, 'labels', 'predictions')
format_submission(df=test, index=index, language_pair="ne-en", method="TransQuest",
path=os.path.join(TEMP_DIRECTORY, SUBMISSION_FILE))
|
