def draw_scatterplot(data_frame, real_column, prediction_column, path, topic):
data_frame = data_frame.sort_values(real_column)
sort_id = list(range(0, len(data_frame.index)))
data_frame['id'] = pd.Series(sort_id).values
data_frame = fit(data_frame, real_column)
data_frame = fit(data_frame, prediction_column)
pearson = pearson_corr(data_frame[real_column].tolist(),
data_frame[prediction_column].tolist())
spearman = spearman_corr(data_frame[real_column].tolist(),
data_frame[prediction_column].tolist())
rmse_value = rmse(data_frame[real_column].tolist(),
data_frame[prediction_column].tolist())
mae = mean_absolute_error(data_frame[real_column].tolist(),
data_frame[prediction_column].tolist())
textstr = 'RMSE=%.4f\nMAE=%.4f\nPearson Correlation=%.4f\nSpearman Correlation=%.4f' % (
rmse_value, mae, pearson, spearman)
plt.figure()
ax = data_frame.plot(kind='scatter',
x='id',
y=real_column,
color='DarkBlue',
label='z_mean',
title=topic)
ax = data_frame.plot(kind='scatter',
x='id',
y=prediction_column,
color='DarkGreen',
label='predicted z_mean',
ax=ax)
ax.text(0.5 * data_frame.shape[0],
min(min(data_frame[real_column].tolist()),
min(data_frame[prediction_column].tolist())),
textstr,
fontsize=10)
fig = ax.get_figure()
fig.savefig(path)
}).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 transformer_config["evaluate_during_training"]:
if transformer_config["n_fold"] > 1:
dev_preds_list = []
test_preds_list = []
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 siamese_transformer_config["evaluate_during_training"]:
if siamese_transformer_config["n_fold"] > 0:
dev_preds = np.zeros((len(dev), siamese_transformer_config["n_fold"]))
test_preds = np.zeros(
(len(test), siamese_transformer_config["n_fold"]))
for i in range(siamese_transformer_config["n_fold"]):
if os.path.exists(
siamese_transformer_config['best_model_dir']
) and os.path.isdir(siamese_transformer_config['best_model_dir']):
shutil.rmtree(siamese_transformer_config['best_model_dir'])
if os.path.exists(