def eval():
if not os.path.exists('./results'):
os.makedirs('./results')
# Load graph
print("Graph loaded")
print("Model name:{}".format(hp.modelname))
# Load data
print("Testing Data...")
txt_src_names, idx_src_names, txt_tgt_names, _ = load_evaluate_data(
eval_mode="test")
x_w2i, x_i2w, y_w2i, y_i2w = load_vocab()
g = Graph(is_training=False)
with g.graph.as_default(), tf.Session() as sess:
sv = tf.train.Saver()
# Restore parameters
print("Parameter Restoring...")
sv.restore(sess,
tf.train.latest_checkpoint(hp.logdir + '/' + hp.modelname))
# Inference
count = 0
with open('./results/' + hp.modelname + '_result.txt', "w") as fout:
for i in range(0, len(txt_src_names), hp.batch_size):
batch_txt_src_names = txt_src_names[i:i + hp.batch_size]
batch_idx_src_names = idx_src_names[i:i + hp.batch_size]
batch_txt_tgt_names = txt_tgt_names[i:i + hp.batch_size]
batch_predicted_ids = sess.run(g.pred_outputs, {
g.x: batch_idx_src_names
}).predicted_ids[:, :, :]
for source, target, predicted_ids in zip(
batch_txt_src_names, batch_txt_tgt_names,
batch_predicted_ids):
print(
str(count) + '\t' + source + '\t' +
hangul.join_jamos(target))
count += 1
candidates = []
predicted_ids = predicted_ids.transpose(1, 0)
for pred in predicted_ids:
candidate = "".join(y_i2w[idx]
for idx in pred).split("E")[0]
candidate = hangul.join_jamos(candidate)
candidates.append(candidate)
fout.write(source + '\t')
fout.write(hangul.join_jamos(target))
for candidate in candidates:
fout.write('\t')
fout.write(candidate.encode('utf-8'))
fout.write('\n')
fout.flush()
def page_text_finder(self, report_text):
page_text = ''
text = ''
found = False
company_name = self.file_nm.split('_')[3]
company_num = self.file_nm.split('_')[4][1:]
company_dict = {'LG상사': 'LG 상사'}
# To resolve hangul encoding issue
company_name = hangul.join_jamos(j2hcj(h2j(company_name)))
if company_name in company_dict.keys():
company_name = company_dict[company_name]
for line in report_text.split('\n'):
if "page_id" in line and '||Title|| ' + company_name in text and company_num in text:
page_text = text
found = True
break
elif "page_id" in line:
text = ''
else:
text += line + '\n'
return page_text, found, company_name, company_num
def save_to_txt(file_nm, file_text):
root_dir = '/Users/daniel/Desktop/test_2/after_inspec_txt/'
path = root_dir + file_nm
path = hangul.join_jamos(j2hcj(h2j(path)))
print(file_nm)
with open(path, 'w') as out_file:
out_file.write(file_text)
def convert_pdf_to_txt(self, pdf_file):
"""PDF파일을 텍스트로 변환해주는 함수
Args:
pdf ([PDF]): PDF파일
Returns:
[dict]: PDF에서 텍스트로 변환된 결과물
"""
output_string = StringIO()
self.file_nm = pdf_file.split(".")[0]
file_ex = pdf_file.split(".")[1]
self.pdf_path = self.report_pdf_dir + pdf_file
self.pdf_path = hangul.join_jamos(j2hcj(h2j(self.pdf_path)))
laparams = LAParams(line_overlap=.5,
char_margin=1.35,
line_margin=1.0,
word_margin=0.01,
boxes_flow=.5,
detect_vertical=False,
all_texts=False)
rsrcmgr = PDFResourceManager()
device = FinanceConverter(rsrcmgr, output_string, laparams=laparams)
interpreter = PDFPageInterpreter(rsrcmgr, device)
# Extract text
found = False
with open(self.pdf_path, 'rb') as in_file:
for page_num, page in enumerate(PDFPage.get_pages(in_file, check_extractable=True)):
interpreter.process_page(page)
page_text = output_string.getvalue()
report_text, found, company_nm, company_num = self.page_text_finder(
page_text)
if found:
break
if not found:
report_text = None
return report_text, company_nm, company_num
def save_to_txt(self, txt):
output_path = self.output_txt_dir + self.file_nm + '.txt'
output_path = hangul.join_jamos(j2hcj(h2j(output_path)))
with open(output_path, 'w') as out_file:
out_file.write(txt)