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),
"English-Latvian-NMT")
print_stat(dev, 'labels', 'predictions')
format_submission(df=test,
index=index,
method="TransQuest",
path=os.path.join(TEMP_DIRECTORY, SUBMISSION_FILE))
train_df, eval_df = train_test_split(train,
test_size=0.1,
random_state=SEED * i)
model = SiameseTransQuestModel(MODEL_NAME,
args=siamesetransquest_config)
model.train_model(train_df, eval_df)
model = SiameseTransQuestModel(
siamesetransquest_config['best_model_dir'])
dev_preds[:, i] = model.predict(dev_sentence_pairs)
test_preds[:, i] = model.predict(test_sentence_pairs)
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))
