File size: 8,378 Bytes
256a159 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 |
import os
from modules.annotator import Annotator
from modules.tokenizer import Tokenizer
import argparse
from collections import Counter
from tqdm import tqdm
import torch
from collections import defaultdict
from multiprocessing import Pool
from opencc import OpenCC
import timeout_decorator
os.environ["TOKENIZERS_PARALLELISM"] = "false"
annotator, sentence_to_tokenized = None, None
cc = OpenCC("t2s")
@timeout_decorator.timeout(10)
def annotate_with_time_out(line):
"""
:param line:
:return:
"""
sent_list = line.split("\t")[1:]
source = sent_list[0]
if args.segmented:
source = source.strip()
else:
source = "".join(source.strip().split())
output_str = ""
for idx, target in enumerate(sent_list[1:]):
try:
if args.segmented:
target = target.strip()
else:
target = "".join(target.strip().split())
if not args.no_simplified:
target = cc.convert(target)
source_tokenized, target_tokenized = sentence_to_tokenized[source], sentence_to_tokenized[target]
out, cors = annotator(source_tokenized, target_tokenized, idx)
if idx == 0:
output_str += "".join(out[:-1])
else:
output_str += "".join(out[1:-1])
except Exception:
raise Exception
return output_str
def annotate(line):
"""
:param line:
:return:
"""
sent_list = line.split("\t")[1:]
source = sent_list[0]
if args.segmented:
source = source.strip()
else:
source = "".join(source.strip().split())
output_str = ""
for idx, target in enumerate(sent_list[1:]):
try:
if args.segmented:
target = target.strip()
else:
target = "".join(target.strip().split())
if not args.no_simplified:
target = cc.convert(target)
source_tokenized, target_tokenized = sentence_to_tokenized[source], sentence_to_tokenized[target]
out, cors = annotator(source_tokenized, target_tokenized, idx)
if idx == 0:
output_str += "".join(out[:-1])
else:
output_str += "".join(out[1:-1])
except Exception:
raise Exception
return output_str
def firsttime_process(args):
tokenizer = Tokenizer(args.granularity, args.device, args.segmented, args.bpe)
global annotator, sentence_to_tokenized
annotator = Annotator.create_default(args.granularity, args.multi_cheapest_strategy)
lines = open(args.file, "r", encoding="utf-8").read().strip().split("\n") # format: id src tgt1 tgt2...
# error_types = []
with open(args.output, "w", encoding="utf-8") as f:
count = 0
sentence_set = set()
sentence_to_tokenized = {}
for line in lines:
sent_list = line.split("\t")[1:]
for idx, sent in enumerate(sent_list):
if args.segmented:
# print(sent)
sent = sent.strip()
else:
sent = "".join(sent.split()).strip()
if idx >= 1:
if not args.no_simplified:
sentence_set.add(cc.convert(sent))
else:
sentence_set.add(sent)
else:
sentence_set.add(sent)
batch = []
for sent in tqdm(sentence_set):
count += 1
if sent:
batch.append(sent)
if count % args.batch_size == 0:
results = tokenizer(batch)
for s, r in zip(batch, results):
sentence_to_tokenized[s] = r # Get tokenization map.
batch = []
if batch:
results = tokenizer(batch)
for s, r in zip(batch, results):
sentence_to_tokenized[s] = r # Get tokenization map.
timeout_indices = []
# 单进程模式
for idx, line in enumerate(tqdm(lines)):
try:
ret = annotate_with_time_out(line)
except Exception:
timeout_indices.append(idx)
return timeout_indices
def main(args):
timeout_indices = firsttime_process(args)
tokenizer = Tokenizer(args.granularity, args.device, args.segmented, args.bpe)
global annotator, sentence_to_tokenized
annotator = Annotator.create_default(args.granularity, args.multi_cheapest_strategy)
lines = open(args.file, "r", encoding="utf-8").read().strip().split("\n")
new_lines = []# format: id src tgt1 tgt2...
with open(args.output, "w", encoding="utf-8") as f:
count = 0
sentence_set = set()
sentence_to_tokenized = {}
for line_idx, line in enumerate(lines):
if line_idx in timeout_indices:
# print(f"line before split: {line}")
line_split = line.split("\t")
line_number, sent_list = line_split[0], line_split[1:]
assert len(sent_list) == 2
sent_list[-1] = " 无"
line = line_number + "\t" + "\t".join(sent_list)
# print(f"line time out: {line}")
new_lines.append(line)
else:
new_lines.append(line)
sent_list = line.split("\t")[1:]
for idx, sent in enumerate(sent_list):
if args.segmented:
# print(sent)
sent = sent.strip()
else:
sent = "".join(sent.split()).strip()
if idx >= 1:
if not args.no_simplified:
sentence_set.add(cc.convert(sent))
else:
sentence_set.add(sent)
else:
sentence_set.add(sent)
batch = []
for sent in tqdm(sentence_set):
count += 1
if sent:
batch.append(sent)
if count % args.batch_size == 0:
results = tokenizer(batch)
for s, r in zip(batch, results):
sentence_to_tokenized[s] = r # Get tokenization map.
batch = []
if batch:
results = tokenizer(batch)
for s, r in zip(batch, results):
sentence_to_tokenized[s] = r # Get tokenization map.
# 单进程模式
lines = new_lines
for idx, line in enumerate(tqdm(lines)):
ret = annotate(line)
f.write(ret)
f.write("\n")
# 多进程模式：仅在Linux环境下测试，建议在linux服务器上使用
# with Pool(args.worker_num) as pool:
# for ret in pool.imap(annotate, tqdm(lines), chunksize=8):
# if ret:
# f.write(ret)
# f.write("\n")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Choose input file to annotate")
parser.add_argument("-f", "--file", type=str, required=True, help="Input parallel file")
parser.add_argument("-o", "--output", type=str, help="Output file", required=True)
parser.add_argument("-b", "--batch_size", type=int, help="The size of batch", default=128)
parser.add_argument("-d", "--device", type=int, help="The ID of GPU", default=0)
parser.add_argument("-w", "--worker_num", type=int, help="The number of workers", default=16)
parser.add_argument("-g", "--granularity", type=str, help="Choose char-level or word-level evaluation", default="char")
parser.add_argument("-m", "--merge", help="Whether merge continuous replacement/deletion/insertion", action="store_true")
parser.add_argument("-s", "--multi_cheapest_strategy", type=str, choices=["first", "all"], default="all")
parser.add_argument("--segmented", help="Whether tokens have been segmented", action="store_true") # 支持提前token化，用空格隔开
parser.add_argument("--no_simplified", help="Whether simplifying chinese", action="store_true") # 将所有corrections转换为简体中文
parser.add_argument("--bpe", help="Whether to use bpe", action="store_true") # 支持 bpe 切分英文单词
args = parser.parse_args()
main(args)
|