def tok():
ori_path = 'training_data/'
in_files = []
out_files = []
for domain in domains:
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_file1_0.4.txt')
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_test1_5k.txt')
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_val1_1k.txt')
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_file2_0.4.txt')
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_test2_5k.txt')
in_files.append(ori_path + 'ori/Twitter_thresholds' +
'/Twitter_110k_val2_1k.txt')
out_files.append(ori_path + 'dif_models_' + domain +
'/source.train.tok')
out_files.append(ori_path + 'dif_models_' + domain +
'/source.test.tok')
out_files.append(ori_path + 'dif_models_' + domain + '/source.val.tok')
out_files.append(ori_path + 'dif_models_' + domain +
'/target.train.tok')
out_files.append(ori_path + 'dif_models_' + domain +
'/target.test.tok')
out_files.append(ori_path + 'dif_models_' + domain + '/target.val.tok')
for f_in, f_out in zip(in_files, out_files):
sens = read_file_lines(f_in)
s_tok = tokenizer(sens, type='word', join=True)
write_file_lines(path=f_out, lines=s_tok)
def tok():
ori_path = '../training_data/'
in_files = []
out_files = []
for domain in domains:
in_files.append(ori_path + 'ori/' + domain + '/train/formal')
in_files.append(ori_path + 'ori/' + domain + '/train/informal')
in_files.append(ori_path + 'ori/' + domain + '/tune/formal.ref0')
in_files.append(ori_path + 'ori/' + domain + '/tune/informal')
in_files.append(ori_path + 'ori/' + domain + '/test/formal.ref0')
in_files.append(ori_path + 'ori/' + domain + '/test/informal')
in_files.append(ori_path + 'ori/' + domain + '/tune/formal')
in_files.append(ori_path + 'ori/' + domain + '/test/formal')
out_files.append(ori_path + 'ori/' + domain + '/train/formal.tok')
out_files.append(ori_path + 'ori/' + domain + '/train/informal.tok')
out_files.append(ori_path + 'ori/' + domain + '/tune/formal.ref0.tok')
out_files.append(ori_path + 'ori/' + domain + '/tune/informal.tok')
out_files.append(ori_path + 'ori/' + domain + '/test/formal.ref0.tok')
out_files.append(ori_path + 'ori/' + domain + '/test/informal.tok')
out_files.append(ori_path + 'ori/' + domain + '/tune/formal.tok')
out_files.append(ori_path + 'ori/' + domain + '/test/formal.tok')
for f_in, f_out in zip(in_files, out_files):
sens = read_file_lines(f_in)
s_tok = tokenizer(sens, type='word', join=True)
write_file_lines(path=f_out, lines=s_tok)
def ensemble_test(domain='fr',
model_type=['ori', 'rule'],
beam_size=4,
max_dec_len=60,
dec_alpha=0.6):
model_dir = [
'./models_' + domain + '/' + t + '/formality_infer/'
for t in model_type
]
input_path = [
'../training_data/dif_models_' + domain + '/eval.' + t
for t in model_type
]
output_path = '../evaluate/gyafc_model_outputs/' + domain + '_out/formal.gpt.' + '_'.join(
model_type) + '.ens'
gpt2 = GPT2(config_path)
generator = ensemble_beam_search_generator(gpt2,
beam_size=beam_size,
model_directorys=model_dir,
max_dec_len=max_dec_len,
dec_alpha=dec_alpha)
sess = generator.build_graph_and_restore(
eos_id=gpt2.text_enc.encode('\n')[0], model_num=len(model_type))
lines = [read_file_lines(p) for p in input_path]
result = []
line_len = [len(l) for l in lines]
max_l, min_l = max(line_len), min(line_len)
assert max_l == min_l
for i in range(0, max_l):
result.append(
generator.generate(
sess, [lines[j][i] for j in range(0, len(model_type))]))
sess.close()
write_file_lines(output_path, result)
def test(model_dir,
input_path,
output_path,
beam_size=4,
max_dec_len=60,
dec_alpha=0.6):
gpt2 = GPT2(config_path)
generator = beam_search_generator(gpt2,
beam_size=beam_size,
model_directory=model_dir,
max_dec_len=max_dec_len,
dec_alpha=dec_alpha)
sess = generator.build_graph_and_restore(
eos_id=gpt2.text_enc.encode('\n')[0])
lines = read_file_lines(input_path)
result = []
for line in lines:
result.append(generator.generate(sess, line))
sess.close()
write_file_lines(output_path, result)
def test(
config_path,
input_num,
model_dir='./models/ori_rule/formality_infer/',
input_path='../training_data/dif_models/eval.ori_rule',
output_path='../evaluate/gyafc_model_outputs/fr_out/formal.gpt.cat_ori_rule.old',
beam_size=4,
max_dec_len=60,
dec_alpha=0.6):
gpt2 = NMT_GPT(config_path=config_path, input_num=input_num)
generator = beam_search_generator(gpt2,
beam_size=beam_size,
model_directory=model_dir,
max_dec_len=max_dec_len,
dec_alpha=dec_alpha)
sess = generator.build_graph_and_restore(
eos_id=gpt2.text_enc.encode('\n')[0])
lines = read_file_lines(input_path)
result = []
for line in lines:
result.append(generator.generate(sess, line, multi_pls=True))
print(line + ' ||| ' + result[-1].strip())
sess.close()
write_file_lines(output_path, result)
def cat_files(f_list, out_path, tokenizer, max_len):
f_lines = [read_file_lines(f) for f in f_list]
new_lines = []
f_len = [len(x) for x in f_lines]
print(f_len)
assert max(f_len) == min(f_len)
length = max(f_len)
for i in range(0, length):
texts = []
for j in range(0, len(f_lines)):
texts.append(f_lines[j][i])
new_lines.append('\t'.join(texts))
return write_file_lines(path=out_path,
lines=new_lines,
tokenizer=tokenizer,
max_len=max_len)