├── __init__.py ├── best_model.pth ├── output └── logs │ └── Experiment_log.log ├── .DS_Store ├── test_predict.py ├── app.py ├── logger.py ├── Readme.md ├── config.py ├── prompt_model.py ├── utils.py ├── predict.py ├── processer.py ├── conlleval.py ├── main.py └── LICENSE /__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /best_model.pth: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /output/logs/Experiment_log.log: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /.DS_Store: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/YuxingLu613/Prompt-NER-Chinese/HEAD/.DS_Store -------------------------------------------------------------------------------- /test_predict.py: -------------------------------------------------------------------------------- 1 | import requests 2 | url = "http://127.0.0.1:5000/predict_ner" 3 | data = { 4 | 'input':'我得了肠胃炎，现在要去做穿肠手术。我昨天吃了二甲双胍，今天准备去拍CT。三个月前我被切除了胃和肾脏，现在觉得特别空虚，希望能做一个白细胞检查。', 5 | } 6 | response = requests.post(url,data=data) 7 | 8 | print(response.text) -------------------------------------------------------------------------------- /app.py: -------------------------------------------------------------------------------- 1 | from multiprocessing.spawn import import_main_path 2 | from flask import Flask,jsonify,request 3 | from pip import main 4 | from predict import build_model, predict 5 | 6 | 7 | app=Flask(__name__) 8 | app.config['JSON_AS_ASCII'] = False 9 | 10 | 11 | @app.route('/') 12 | def hello(): 13 | return 'Hello World!' 14 | 15 | @app.route('/predict_ner',methods=["POST"]) 16 | def predict_ner(): 17 | global model 18 | input=request.form.get("input") 19 | output=predict(input,model) 20 | return jsonify(output=output) 21 | 22 | if __name__=="__main__": 23 | model=build_model() 24 | app.run(host="127.0.0.1") -------------------------------------------------------------------------------- /logger.py: -------------------------------------------------------------------------------- 1 | import logging 2 | import os 3 | from config import Config 4 | 5 | config=Config() 6 | 7 | if not os.path.exists(config.log_path): 8 | os.makedirs(config.log_path) 9 | 10 | logger=logging.getLogger("client_log") 11 | 12 | logger.setLevel(logging.INFO) 13 | 14 | stream_handler=logging.StreamHandler() 15 | log_file_handler=logging.FileHandler(filename=os.path.join(config.log_path,config.log_name),encoding="utf-8") 16 | 17 | formatter = logging.Formatter("%(asctime)s - %(filename)s[line:%(lineno)d] - %(message)s") 18 | 19 | stream_handler.setFormatter(formatter) 20 | log_file_handler.setFormatter(formatter) 21 | 22 | logger.addHandler(stream_handler) 23 | logger.addHandler(log_file_handler) -------------------------------------------------------------------------------- /Readme.md: -------------------------------------------------------------------------------- 1 | # Prompt-NER on Chinese EHR data 2 | 3 | ### 简介 4 | 5 | 该仓库使用[BERT](https://github.com/google-research/bert)作为预训练模型使用Prompt预训练方法进行命名实体识别任务。 6 | 7 | 8 | 9 | ### 文件夹介绍： 10 | 11 | ``` 12 | . 13 | ├── .DS_Store 14 | ├── __init__.py 15 | ├── app.py # 接口文件 16 | ├── best_model.pth # 模型文件（需要自己训练） 17 | ├── config.py # 配置文件 18 | ├── conlleval.py # 评价指标 19 | ├── data # 数据集 20 | │   ├── eval.txt # 处理好的验证集 21 | │   ├── test.txt # 处理好的测试集 22 | │   └── train.txt # 处理好的训练集 23 | ├── logger.py # 日志文件 24 | ├── main.py # 主文件 25 | ├── output # 输出 26 | │   └── logs 27 | │   └── Experiment_log.log 28 | ├── predict.py # 预测程序 29 | ├── processer.py # 数据预处理文件 30 | ├── prompt_model.py # 模型结构 31 | ├── test_predict.py # 接口测试文件 32 | └── utils.py # 方法函数 33 | ``` 34 | 35 | 36 | 37 | ### Requirements: 38 | 39 | python 40 | 41 | torch 42 | 43 | sklearn 44 | 45 | pandas 46 | 47 | transformers 48 | 49 | 50 | 51 | ### 直接使用方法： 52 | 53 | 使用python运行main.py，获得模型文件。 54 | 55 | 使用python运行app.py，并且在test.py中修改input数据，获得返回的结果。 56 | 57 | 58 | 59 | ### 结果： 60 | 61 | | 训练方法 | F1 | 62 | | -------------- | ------ | 63 | | 常规预训练方法 | 0.7617 | 64 | | Prompt训练方法 | 0.8189 | -------------------------------------------------------------------------------- /config.py: -------------------------------------------------------------------------------- 1 | import datetime 2 | import os 3 | import threading 4 | 5 | 6 | class Config(object): 7 | _instance_lock = threading.Lock() 8 | _init_flag = False 9 | 10 | def __init__(self): 11 | if not Config._init_flag: 12 | Config._init_flag = True 13 | root_path = str(os.getcwd()).replace("\\", "/") 14 | if 'source' in root_path.split('/'): 15 | self.base_path = os.path.abspath(os.path.join(os.path.pardir)) 16 | else: 17 | self.base_path = os.path.abspath(os.path.join(os.getcwd())) 18 | self._init_train_config() 19 | 20 | def __new__(cls, *args, **kwargs): 21 | """ 22 | 单例类 23 | :param args: 24 | :param kwargs: 25 | :return: 26 | """ 27 | if not hasattr(Config, '_instance'): 28 | with Config._instance_lock: 29 | if not hasattr(Config, '_instance'): 30 | Config._instance = object.__new__(cls) 31 | return Config._instance 32 | 33 | def _init_train_config(self): 34 | self.label_list = ["O","B-手术","I-手术","B-药物","I-药物","B-实验室检验","I-实验室检验","B-影像检查","I-影像检查","B-解剖部位","I-解剖部位","B-疾病和诊断","I-疾病和诊断"] 35 | self.use_gpu = True 36 | self.device = "cpu" 37 | self.sep = " " 38 | 39 | 40 | # 输入数据集、输出目录 41 | self.train_file = os.path.join(self.base_path, 'data', 'train.txt') 42 | self.eval_file = os.path.join(self.base_path, 'data', 'eval.txt') 43 | self.test_file = os.path.join(self.base_path, 'data', 'test.txt') 44 | self.log_path = os.path.join(self.base_path, 'output', "logs") 45 | self.log_name="Experiment_log.log" 46 | self.output_path = os.path.join(self.base_path, 'output', datetime.datetime.now().strftime('%Y%m%d%H%M%S')) 47 | 48 | model_list=["bert-base-chinese","hfl/chinese-bert-wwm-ext","hfl/chinese-roberta-wwm-ext","voidful/albert_chinese_base","hfl/chinese-electra-180g-base-discriminator","nghuyong/ernie-1.0"] 49 | 50 | # Pretrained model name or path if not the same as model_name 51 | # self.model_name_or_path = "/home/luyx/PLM/Continue_Pretrain/training_output" 52 | self.model_name_or_path = model_list[0] 53 | 54 | # 以下是模型训练参数 55 | self.do_train = True 56 | self.do_eval = True 57 | self.do_test = True 58 | self.clean = True 59 | self.need_birnn = True 60 | self.do_lower_case = True 61 | self.rnn_dim = 768 62 | self.max_seq_length = 64 63 | self.train_batch_size = 3072 64 | self.eval_batch_size = 3072 65 | self.num_train_epochs = 30 66 | self.gradient_accumulation_steps = 1 67 | self.learning_rate = 5e-6 68 | self.adam_epsilon = 1e-8 69 | self.warmup_steps = 0 70 | self.logging_steps = 100 71 | self.train_eval_split=0.8 72 | self.min_sequence_length=10 -------------------------------------------------------------------------------- /prompt_model.py: -------------------------------------------------------------------------------- 1 | from re import sub 2 | from select import select 3 | from transformers import BertForMaskedLM,AutoConfig,AutoModel,AutoModelForMaskedLM,BertForTokenClassification 4 | import torch 5 | from torch import nn 6 | from torch.utils.data import DataLoader, SequentialSampler 7 | from tqdm import tqdm 8 | import conlleval 9 | 10 | class Prompt_Based_NER(nn.Module): 11 | 12 | def __init__(self, config): 13 | super(Prompt_Based_NER, self).__init__() 14 | self.config=config 15 | self.plm_config=AutoConfig.from_pretrained(self.config.model_name_or_path) 16 | # self.plm_model=BertForMaskedLM.from_pretrained("bert-base-chinese") 17 | self.plm_model=BertForMaskedLM.from_pretrained(self.config.model_name_or_path) 18 | self.dropout=nn.Dropout(self.plm_config.hidden_dropout_prob) 19 | self.linear1=nn.Linear(in_features=self.plm_config.vocab_size,out_features=256) 20 | self.linear2=nn.Linear(in_features=256,out_features=len(self.config.label_list)) 21 | self.softmax=nn.Softmax(dim=-1) 22 | 23 | def forward(self, input_ids, token_type_ids=None, attention_mask=None,masked_index=None): 24 | masked_index=masked_index.to(self.config.device) 25 | output=self.plm_model(input_ids, token_type_ids=token_type_ids,attention_mask=attention_mask)[0] 26 | selected_output=None 27 | for index, sub_output in enumerate(output): 28 | if selected_output is None: 29 | selected_output=sub_output[masked_index[index]].unsqueeze(0) 30 | else: 31 | selected_output=torch.cat((selected_output,sub_output[masked_index[index]].unsqueeze(0)),dim=0) 32 | output=selected_output 33 | output=self.linear1(output) 34 | output=self.dropout(output) 35 | output=self.linear2(output) 36 | return output 37 | 38 | def predict(self, input_ids, token_type_ids=None, attention_mask=None): 39 | output=self.plm_model(input_ids, token_type_ids=token_type_ids,attention_mask=attention_mask).logits 40 | output=self.hidden2tag(output) 41 | output=torch.argmax(output,dim=-1) 42 | return output 43 | 44 | def evaluate(config, dataloader, model, id2label, all_ori_tokens): 45 | ori_labels, pred_labels = [], [] 46 | model.eval() 47 | 48 | for b_i, (input_ids, token_type_ids, attention_mask, label_ids,ori_tokens,masked_index) in enumerate(tqdm(dataloader, desc="Evaluating")): 49 | input_ids = input_ids.to(config.device) 50 | attention_mask = attention_mask.to(config.device) 51 | token_type_ids = token_type_ids.to(config.device) 52 | label_ids = label_ids.to(config.device) 53 | masked_index=masked_index.to(config.device) 54 | with torch.no_grad(): 55 | output = model.forward(input_ids, token_type_ids, attention_mask,masked_index) 56 | logits=torch.argmax(output,dim=-1) 57 | 58 | for l in logits: 59 | pred_labels.append([id2label[l.item()]]) 60 | 61 | for l in label_ids: 62 | ori_labels.append([id2label[l.tolist().index(1)]]) 63 | 64 | eval_list = [] 65 | for ori_tokens, oril, prel in zip(all_ori_tokens, ori_labels, pred_labels): 66 | for ot, ol, pl in zip(ori_tokens, oril, prel): 67 | if ot in ["[CLS]", "[SEP]"]: 68 | continue 69 | eval_list.append(f"{ot} {ol} {pl}\n") 70 | eval_list.append("\n") 71 | 72 | # eval the model 73 | counts = conlleval.evaluate(eval_list) 74 | conlleval.report(counts) 75 | 76 | # namedtuple('Metrics', 'tp fp fn prec rec fscore') 77 | overall, by_type = conlleval.metrics(counts) 78 | return overall, by_type -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- 1 | import csv 2 | import json 3 | import os 4 | import pickle 5 | from re import I 6 | import time 7 | from datetime import timedelta,datetime 8 | import numpy as np 9 | import pandas as pd 10 | from sklearn.model_selection import train_test_split 11 | 12 | 13 | def load_file(file_path: str, separater: str = None): 14 | """ 15 | 读取文件； 16 | 若sep为None，按行读取，返回文件内容列表，格式为:[xxx,xxx,xxx,...] 17 | 若不为None，按行读取分隔，返回文件内容列表，格式为: [[xxx,xxx],[xxx,xxx],...] 18 | :param filepath: 19 | :param sep: 20 | :return: 21 | """ 22 | with open(file_path,"r",encoding="utf-8-sig") as f: 23 | lines=f.readlines() 24 | if separater: 25 | return [line.strip().split(separater) for line in lines] 26 | else: 27 | return lines 28 | 29 | 30 | def load_csv(file_path: str, is_tsv: bool = False): 31 | """ 32 | 加载csv文件为OrderDict()列表 33 | :param filepath: 34 | :param is_tsv: 35 | :return: 36 | """ 37 | dialect = 'excel-tab' if is_tsv else 'excel' 38 | with open(file_path, "r", encoding='utf-8') as f: 39 | reader = csv.DictReader(f, dialect=dialect) 40 | return list(reader) 41 | 42 | def save_csv(data: pd.DataFrame, file_path: str,sep: str=","): 43 | """ 44 | 将DataFrame数据保存到对应file_path下 45 | :param data: 46 | :para file_path: 47 | :return: 48 | """ 49 | data.to_csv(file_path,sep=sep) 50 | 51 | 52 | def load_json(filepath: str): 53 | """ 54 | 加载json文件 55 | :param filepath: 56 | :return: 57 | """ 58 | with open(filepath, "r", encoding="utf-8") as f: 59 | return [json.loads(line.strip(), encoding="utf-8") for line in f.readlines()] 60 | 61 | 62 | def save_json(list_data, filepath): 63 | """ 64 | 保存json文件 65 | :param list_data: 66 | :param filepath: 67 | :return: 68 | """ 69 | with open(filepath, "w", encoding="utf-8") as f: 70 | for data in list_data: 71 | json_str = json.dumps(data, ensure_ascii=False) 72 | f.write("{}\n".format(json_str)) 73 | f.flush() 74 | 75 | 76 | def load_pkl(filepath): 77 | """ 78 | 加载pkl文件 79 | :param filepath: 80 | :return: 81 | """ 82 | with open(filepath, 'rb') as f: 83 | data = pickle.load(f) 84 | return data 85 | 86 | 87 | def save_pkl(data, filepath): 88 | """ 89 | 保存pkl文件，数据序列化 90 | :param data: 91 | :param filepath: 92 | :return: 93 | """ 94 | with open(filepath, 'wb') as f: 95 | pickle.dump(data, f) 96 | 97 | 98 | def calculate_distance(vector1, vector2,mode="cos"): 99 | """ 100 | 计算两个向量的余弦相似度 101 | :param vector1: 向量1 102 | :param vector2: 向量2 103 | :param mode: "cos"余弦相似度，"euc"欧氏距离 104 | :return: 105 | """ 106 | if mode=="cos": 107 | distance = np.dot(vector1, vector2) / (np.linalg.norm(vector1) * (np.linalg.norm(vector2))) # 余弦夹角 108 | elif mode=="euc": 109 | distance = np.sqrt(np.sum(np.square(vector1 - vector2))) # 欧式距离 110 | return distance 111 | 112 | 113 | def split_data(data_set, ratio): 114 | """ 115 | 数据集切分 116 | :param data_set: 117 | :param ratio:分割比例 118 | :return: 119 | """ 120 | split_data1,split_data2=train_test_split(data_set,ratio) 121 | return split_data1,split_data2 122 | 123 | 124 | def format_data(t: datetime): 125 | """ 126 | 时间格式化，time.strftime("%Y-%m-%d %H:%M:%S") 127 | :param t: 128 | :return: 129 | """ 130 | return t.strftime("%Y-%m-%d %H:%M:%S") 131 | 132 | 133 | def get_used_dif(start_time): 134 | """ 135 | 获取已使用时间 136 | :param start_time: time.time() 137 | :return: 138 | """ 139 | end_time = time.time() 140 | time_dif = end_time - start_time 141 | return timedelta(seconds=int(round(time_dif))) 142 | 143 | 144 | def scan_filepath(path): 145 | """ 146 | 递归返回指定目录下的所有文件 147 | :param ph: 148 | :return: 149 | """ 150 | path_list = [] 151 | for p in os.listdir(path): 152 | fp = os.path.join(path, p) 153 | if os.path.isfile(fp): 154 | path_list.append(fp) 155 | elif os.path.isdir(fp): 156 | path_list.extend(scan_filepath(fp)) 157 | return path_list 158 | -------------------------------------------------------------------------------- /predict.py: -------------------------------------------------------------------------------- 1 | from prompt_model import Prompt_Based_NER 2 | import torch 3 | from torch import nn 4 | from config import Config 5 | import os 6 | from processer import get_label2id_id2label 7 | from transformers import BertTokenizer 8 | from tqdm import tqdm 9 | from torch.utils.data import DataLoader,SequentialSampler,TensorDataset 10 | from prompt_model import Prompt_Based_NER 11 | 12 | def build_model(): 13 | config=Config() 14 | # model_path=os.path.join(config.base_path,"output","20220306205222","best_model.pth") 15 | model=Prompt_Based_NER(config) 16 | # net_dict = model.state_dict() 17 | # best_model = torch.load(model_path) 18 | # for k, v in best_model.items(): 19 | # name = k[7:] # remove `module.` 20 | # net_dict[name] = v 21 | # model.load_state_dict(net_dict) 22 | # model.to(device) 23 | # torch.save(model.state_dict(),"best_model.pth") 24 | 25 | model.load_state_dict(torch.load("best_model.pth")) 26 | 27 | return model 28 | 29 | def predict(input,model): 30 | config=Config() 31 | 32 | use_gpu = False 33 | device = torch.device('cuda' if use_gpu else config.device) 34 | config.device = device 35 | 36 | tokenizer = BertTokenizer.from_pretrained(config.model_name_or_path) 37 | 38 | label2id, id2label = get_label2id_id2label(config.output_path, label_list=config.label_list) 39 | 40 | def get_input_sequence(input): 41 | input_split=input.split("。") 42 | input_list=[] 43 | max_len=config.max_seq_length-6 44 | for input in input_split: 45 | if input: 46 | input+="。" 47 | while len(input)>0: 48 | input_list.append(input[:max_len]) 49 | input=input[max_len:] 50 | return input_list 51 | 52 | class Template(): 53 | def __init__(self,text) -> None: 54 | self.text=text 55 | 56 | def create_template(input_list): 57 | dataset=[] 58 | for text_list in input_list: 59 | for word in list(text_list): 60 | template=Template(text=text_list+"，"+word+"是") 61 | dataset.append(template) 62 | return dataset 63 | 64 | class InputFeatures(object): 65 | """A single set of features of data.""" 66 | 67 | def __init__(self, input_ids, token_type_ids, attention_mask, masked_index): 68 | """ 69 | :param input_ids: 单词在词典中的编码 70 | :param attention_mask: 指定 对哪些词 进行self-Attention操作 71 | :param token_type_ids: 区分两个句子的编码（上句全为0，下句全为1） 72 | :param label_id: 标签的id 73 | """ 74 | self.input_ids = input_ids 75 | self.token_type_ids = token_type_ids 76 | self.attention_mask = attention_mask 77 | self.masked_index = masked_index 78 | 79 | def convert_examples_to_features(dataset,tokenizer): 80 | features=[] 81 | max_seq_length = config.max_seq_length 82 | for example in tqdm(dataset): 83 | if len(example.text) >= max_seq_length - 3: 84 | # -2的原因是因为序列需要加一个句首和句尾标志 85 | example.text = example.text[0:(max_seq_length - 3)] 86 | example.text = ["[CLS"]+list(example.text) + ['[MASK]']+["[SEP]"] 87 | 88 | input_ids = tokenizer.convert_tokens_to_ids(example.text) 89 | attention_mask = [1]*len(example.text) 90 | masked_index = example.text.index("[MASK]") 91 | 92 | if len(input_ids) < max_seq_length: 93 | input_ids.extend([0]*(max_seq_length-len(input_ids))) 94 | attention_mask.extend([0]*(max_seq_length-len(attention_mask))) 95 | 96 | token_type_ids = [0]*max_seq_length 97 | 98 | assert(len(input_ids) == len(attention_mask) == 99 | len(token_type_ids) == max_seq_length) 100 | 101 | features.append(InputFeatures(input_ids=input_ids, 102 | token_type_ids=token_type_ids, 103 | attention_mask=attention_mask, 104 | masked_index=masked_index)) 105 | 106 | all_input_ids = torch.tensor( 107 | [f.input_ids for f in features], dtype=torch.long) 108 | all_token_type_ids = torch.tensor( 109 | [f.token_type_ids for f in features], dtype=torch.long) 110 | all_attention_mask = torch.tensor( 111 | [f.attention_mask for f in features], dtype=torch.long) 112 | all_masked_index = torch.tensor( 113 | [f.masked_index for f in features], dtype=torch.long) 114 | data = TensorDataset(all_input_ids, all_token_type_ids, 115 | all_attention_mask, all_masked_index) 116 | 117 | return features, data 118 | 119 | input_list=get_input_sequence(input) 120 | dataset=create_template(input_list) 121 | 122 | predict_features, predict_dataset=convert_examples_to_features(dataset,tokenizer) 123 | predict_dataloader=DataLoader(predict_dataset,batch_size=config.eval_batch_size,sampler=SequentialSampler(predict_dataset)) 124 | 125 | model.eval() 126 | 127 | pred_labels = [] 128 | 129 | for b_i, (input_ids, token_type_ids, attention_mask,masked_index) in enumerate(tqdm(predict_dataloader, desc="Predicting")): 130 | input_ids = input_ids.to(config.device) 131 | attention_mask = attention_mask.to(config.device) 132 | token_type_ids = token_type_ids.to(config.device) 133 | masked_index=masked_index.to(config.device) 134 | with torch.no_grad(): 135 | output = model.forward(input_ids, token_type_ids, attention_mask,masked_index) 136 | logits=torch.argmax(output,dim=-1) 137 | 138 | for l in logits: 139 | pred_labels.append([id2label[l.item()]]) 140 | 141 | output=[] 142 | 143 | for word,label in zip(list("".join(input_list)),pred_labels): 144 | output.append((word,label[0])) 145 | print(output) 146 | return output 147 | 148 | 149 | if __name__=="__main__": 150 | model=build_model() 151 | input="我得了肠胃炎，现在要去做穿肠手术。我昨天吃了二甲双胍，今天准备去拍CT。三个月前我被切除了胃和肾脏，现在觉得特别空虚，希望能做一个白细胞检查。" 152 | for word,label in predict(input,model): 153 | print(word,label) 154 | 155 | 156 | -------------------------------------------------------------------------------- /processer.py: -------------------------------------------------------------------------------- 1 | import enum 2 | import torch 3 | from utils import load_file, save_pkl, load_pkl 4 | from config import Config 5 | from logger import logger as logging 6 | import os 7 | from tqdm import trange, tqdm 8 | from torch.utils.data import TensorDataset 9 | 10 | config = Config() 11 | 12 | 13 | def dataset_format(file_path, separater=" "): 14 | label_map = {label: i for i, label in enumerate(config.label_list)} 15 | sentence_list = [] 16 | label_list = [] 17 | sentence = "" 18 | labels = [] 19 | file = load_file(file_path, separater=separater) 20 | for unit in file[:]: 21 | if len(unit) <= 1: 22 | if len(sentence) >= config.min_sequence_length: 23 | sentence_list.append(sentence[:58]) 24 | sentence = "" 25 | label_list.append(labels[:58]) 26 | labels = [] 27 | continue 28 | word, label = unit 29 | if word == "，" or word == "。": 30 | if len(sentence) >= config.min_sequence_length: 31 | sentence_list.append(sentence[:58]) 32 | sentence = "" 33 | label_list.append(labels[:58]) 34 | labels = [] 35 | else: 36 | sentence += word 37 | labels.append(label_map[label]) 38 | return sentence_list, label_list 39 | 40 | 41 | class Template(): 42 | def __init__(self, text, label) -> None: 43 | self.text = text 44 | self.label = label 45 | 46 | 47 | def create_template(all_texts, all_labels): 48 | dataset = [] 49 | for item in zip(all_texts, all_labels): 50 | texts, labels = item 51 | for word, label in zip(list(texts), labels): 52 | template = Template(text=texts+"，"+word+"是", label=label) 53 | dataset.append(template) 54 | return dataset 55 | 56 | 57 | class InputFeatures(object): 58 | """A single set of features of data.""" 59 | 60 | def __init__(self, input_ids, token_type_ids, attention_mask, masked_index, label_id, ori_label): 61 | """ 62 | :param input_ids: 单词在词典中的编码 63 | :param attention_mask: 指定 对哪些词 进行self-Attention操作 64 | :param token_type_ids: 区分两个句子的编码（上句全为0，下句全为1） 65 | :param label_id: 标签的id 66 | """ 67 | self.input_ids = input_ids 68 | self.token_type_ids = token_type_ids 69 | self.attention_mask = attention_mask 70 | self.label_id = label_id 71 | self.ori_label = ori_label 72 | self.masked_index = masked_index 73 | 74 | 75 | def convert_examples_to_features(dataset, tokenizer): 76 | features = [] 77 | label_map = {label: i for i, label in enumerate(config.label_list)} 78 | max_seq_length = config.max_seq_length 79 | for example in tqdm(dataset): 80 | 81 | label = [0]*len(config.label_list) 82 | label[example.label] = 1 83 | 84 | # for i, word in enumerate(example_text): 85 | # token = tokenizer.tokenize(word) 86 | # tokens.extend(token) 87 | # ori_tokens.append(word) 88 | 89 | if len(example.text) >= max_seq_length - 8: 90 | # -2的原因是因为序列需要加一个句首和句尾标志 91 | example.text = example.text[0:(max_seq_length - 9)] 92 | 93 | example.text = ["[CLS"]+list(example.text) + ['[MASK]']+["[SEP]"] 94 | 95 | input_ids = tokenizer.convert_tokens_to_ids(example.text) 96 | attention_mask = [1]*len(example.text) 97 | masked_index = example.text.index("[MASK]") 98 | 99 | if len(input_ids) < max_seq_length: 100 | input_ids.extend([0]*(max_seq_length-len(input_ids))) 101 | attention_mask.extend([0]*(max_seq_length-len(attention_mask))) 102 | 103 | token_type_ids = [0]*max_seq_length 104 | 105 | assert(len(input_ids) == len(attention_mask) == 106 | len(token_type_ids) == max_seq_length) 107 | 108 | features.append(InputFeatures(input_ids=input_ids, 109 | token_type_ids=token_type_ids, 110 | attention_mask=attention_mask, 111 | label_id=label, 112 | ori_label=example.label, 113 | masked_index=masked_index)) 114 | 115 | all_input_ids = torch.tensor( 116 | [f.input_ids for f in features], dtype=torch.long) 117 | all_token_type_ids = torch.tensor( 118 | [f.token_type_ids for f in features], dtype=torch.long) 119 | all_attention_mask = torch.tensor( 120 | [f.attention_mask for f in features], dtype=torch.long) 121 | all_label_ids = torch.tensor( 122 | [f.label_id for f in features], dtype=torch.float) 123 | all_ori_labels = torch.tensor( 124 | [f.ori_label for f in features], dtype=torch.long) 125 | all_masked_index = torch.tensor( 126 | [f.masked_index for f in features], dtype=torch.long) 127 | data = TensorDataset(all_input_ids, all_token_type_ids, 128 | all_attention_mask, all_label_ids, all_ori_labels, all_masked_index) 129 | return features, data 130 | 131 | 132 | def get_labels(config: Config): 133 | """ 134 | 读取训练数据获取标签 135 | :param config: 136 | :return: 137 | """ 138 | label_map = {label: i for i, label in enumerate(config.label_list)} 139 | label_pkl_path = "label_list.pkl" 140 | save_pkl(label_map, label_pkl_path) 141 | if os.path.exists(label_pkl_path): 142 | logging.info(f"loading labels info from {label_pkl_path}") 143 | labels = load_pkl(label_pkl_path) 144 | print(labels) 145 | else: 146 | logging.info( 147 | f"loading labels info from train file and dump in {config.output_path}") 148 | tokens_list = load_file(config.train_file, separater=config.sep) 149 | labels = set([tokens[1] for tokens in tokens_list if len(tokens) == 2]) 150 | 151 | if len(labels) == 0: 152 | ValueError("loading labels error, labels type not found in data file: {}".format( 153 | config.output_path)) 154 | else: 155 | save_pkl(labels, label_pkl_path) 156 | 157 | return labels 158 | 159 | 160 | def get_label2id_id2label(output_path, label_list): 161 | """ 162 | 获取label2id、id2label的映射 163 | :param output_path: 164 | :param label_list: 165 | :return: 166 | """ 167 | label2id_path = "label2id.pkl" 168 | if os.path.exists(label2id_path): 169 | label2id = load_pkl(label2id_path) 170 | else: 171 | label2id = {l: i for i, l in enumerate(label_list)} 172 | save_pkl(label2id, label2id_path) 173 | 174 | id2label = {value: key for key, value in label2id.items()} 175 | return label2id, id2label 176 | -------------------------------------------------------------------------------- /conlleval.py: -------------------------------------------------------------------------------- 1 | # Python version of the evaluation script from CoNLL'00- 2 | # Originates from: https://github.com/spyysalo/conlleval.py 3 | 4 | 5 | # Intentional differences: 6 | # - accept any space as delimiter by default 7 | # - optional file argument (default STDIN) 8 | # - option to set boundary (-b argument) 9 | # - LaTeX output (-l argument) not supported 10 | # - raw tags (-r argument) not supported 11 | 12 | # add function :evaluate(predicted_label, ori_label): which will not read from file 13 | 14 | import codecs 15 | import re 16 | import sys 17 | from collections import defaultdict, namedtuple 18 | 19 | ANY_SPACE = '' 20 | 21 | 22 | class FormatError(Exception): 23 | pass 24 | 25 | 26 | Metrics = namedtuple('Metrics', 'tp fp fn prec rec fscore') 27 | 28 | 29 | class EvalCounts(object): 30 | def __init__(self): 31 | self.correct_chunk = 0 # number of correctly identified chunks 32 | self.correct_tags = 0 # number of correct chunk tags 33 | self.found_correct = 0 # number of chunks in corpus 34 | self.found_guessed = 0 # number of identified chunks 35 | self.token_counter = 0 # token counter (ignores sentence breaks) 36 | 37 | # counts by type 38 | self.t_correct_chunk = defaultdict(int) 39 | self.t_found_correct = defaultdict(int) 40 | self.t_found_guessed = defaultdict(int) 41 | 42 | 43 | def parse_args(argv): 44 | import argparse 45 | parser = argparse.ArgumentParser( 46 | description='evaluate tagging results using CoNLL criteria', 47 | formatter_class=argparse.ArgumentDefaultsHelpFormatter 48 | ) 49 | arg = parser.add_argument 50 | arg('-b', '--boundary', metavar='STR', default='-X-', 51 | help='sentence boundary') 52 | arg('-d', '--delimiter', metavar='CHAR', default=ANY_SPACE, 53 | help='character delimiting items in input') 54 | arg('-o', '--otag', metavar='CHAR', default='O', 55 | help='alternative outside tag') 56 | arg('file', nargs='?', default=None) 57 | return parser.parse_args(argv) 58 | 59 | 60 | def parse_tag(t): 61 | m = re.match(r'^([^-]*)-(.*)$', t) 62 | return m.groups() if m else (t, '') 63 | 64 | 65 | def evaluate(iterable, options=None): 66 | if options is None: 67 | options = parse_args([]) # use defaults 68 | 69 | counts = EvalCounts() 70 | num_features = None # number of features per line 71 | in_correct = False # currently processed chunks is correct until now 72 | last_correct = 'O' # previous chunk tag in corpus 73 | last_correct_type = '' # type of previously identified chunk tag 74 | last_guessed = 'O' # previously identified chunk tag 75 | last_guessed_type = '' # type of previous chunk tag in corpus 76 | 77 | for line in iterable: 78 | line = line.rstrip('\r\n') 79 | 80 | if options.delimiter == ANY_SPACE: 81 | features = line.split() 82 | else: 83 | features = line.split(options.delimiter) 84 | 85 | if num_features is None: 86 | num_features = len(features) 87 | elif num_features != len(features) and len(features) != 0: 88 | raise FormatError('unexpected number of features: %d (%d)' % 89 | (len(features), num_features)) 90 | 91 | if len(features) == 0 or features[0] == options.boundary: 92 | features = [options.boundary, 'O', 'O'] 93 | if len(features) < 3: 94 | raise FormatError('unexpected number of features in line %s' % line) 95 | 96 | guessed, guessed_type = parse_tag(features.pop()) 97 | correct, correct_type = parse_tag(features.pop()) 98 | first_item = features.pop(0) 99 | 100 | if first_item == options.boundary: 101 | guessed = 'O' 102 | 103 | end_correct = end_of_chunk(last_correct, correct, 104 | last_correct_type, correct_type) 105 | end_guessed = end_of_chunk(last_guessed, guessed, 106 | last_guessed_type, guessed_type) 107 | start_correct = start_of_chunk(last_correct, correct, 108 | last_correct_type, correct_type) 109 | start_guessed = start_of_chunk(last_guessed, guessed, 110 | last_guessed_type, guessed_type) 111 | 112 | if in_correct: 113 | if (end_correct and end_guessed and 114 | last_guessed_type == last_correct_type): 115 | in_correct = False 116 | counts.correct_chunk += 1 117 | counts.t_correct_chunk[last_correct_type] += 1 118 | elif (end_correct != end_guessed or guessed_type != correct_type): 119 | in_correct = False 120 | 121 | if start_correct and start_guessed and guessed_type == correct_type: 122 | in_correct = True 123 | 124 | if start_correct: 125 | counts.found_correct += 1 126 | counts.t_found_correct[correct_type] += 1 127 | if start_guessed: 128 | counts.found_guessed += 1 129 | counts.t_found_guessed[guessed_type] += 1 130 | if first_item != options.boundary: 131 | if correct == guessed and guessed_type == correct_type: 132 | counts.correct_tags += 1 133 | counts.token_counter += 1 134 | 135 | last_guessed = guessed 136 | last_correct = correct 137 | last_guessed_type = guessed_type 138 | last_correct_type = correct_type 139 | 140 | if in_correct: 141 | counts.correct_chunk += 1 142 | counts.t_correct_chunk[last_correct_type] += 1 143 | 144 | return counts 145 | 146 | 147 | def uniq(iterable): 148 | seen = set() 149 | return [i for i in iterable if not (i in seen or seen.add(i))] 150 | 151 | 152 | def calculate_metrics(correct, guessed, total): 153 | tp, fp, fn = correct, guessed - correct, total - correct 154 | p = 0 if tp + fp == 0 else 1. * tp / (tp + fp) 155 | r = 0 if tp + fn == 0 else 1. * tp / (tp + fn) 156 | f = 0 if p + r == 0 else 2 * p * r / (p + r) 157 | return Metrics(tp, fp, fn, p, r, f) 158 | 159 | 160 | def metrics(counts): 161 | c = counts 162 | overall = calculate_metrics( 163 | c.correct_chunk, c.found_guessed, c.found_correct 164 | ) 165 | by_type = {} 166 | for t in uniq(list(c.t_found_correct) + list(c.t_found_guessed)): 167 | by_type[t] = calculate_metrics( 168 | c.t_correct_chunk[t], c.t_found_guessed[t], c.t_found_correct[t] 169 | ) 170 | return overall, by_type 171 | 172 | 173 | def report(counts, out=None): 174 | if out is None: 175 | out = sys.stdout 176 | 177 | overall, by_type = metrics(counts) 178 | 179 | c = counts 180 | out.write('processed %d tokens with %d phrases; ' % 181 | (c.token_counter, c.found_correct)) 182 | out.write('found: %d phrases; correct: %d.\n' % 183 | (c.found_guessed, c.correct_chunk)) 184 | 185 | if c.token_counter > 0: 186 | out.write('accuracy: %6.2f%%; ' % 187 | (100. * c.correct_tags / c.token_counter)) 188 | out.write('precision: %6.2f%%; ' % (100. * overall.prec)) 189 | out.write('recall: %6.2f%%; ' % (100. * overall.rec)) 190 | out.write('FB1: %6.2f\n' % (100. * overall.fscore)) 191 | 192 | for i, m in sorted(by_type.items()): 193 | out.write('%17s: ' % i) 194 | out.write('precision: %6.2f%%; ' % (100. * m.prec)) 195 | out.write('recall: %6.2f%%; ' % (100. * m.rec)) 196 | out.write('FB1: %6.2f %d\n' % (100. * m.fscore, c.t_found_guessed[i])) 197 | 198 | 199 | def report_notprint(counts, out=None): 200 | if out is None: 201 | out = sys.stdout 202 | 203 | overall, by_type = metrics(counts) 204 | 205 | c = counts 206 | final_report = [] 207 | line = [] 208 | line.append('processed %d tokens with %d phrases; ' % 209 | (c.token_counter, c.found_correct)) 210 | line.append('found: %d phrases; correct: %d.\n' % 211 | (c.found_guessed, c.correct_chunk)) 212 | final_report.append("".join(line)) 213 | 214 | if c.token_counter > 0: 215 | line = [] 216 | line.append('accuracy: %6.2f%%; ' % 217 | (100. * c.correct_tags / c.token_counter)) 218 | line.append('precision: %6.2f%%; ' % (100. * overall.prec)) 219 | line.append('recall: %6.2f%%; ' % (100. * overall.rec)) 220 | line.append('FB1: %6.2f\n' % (100. * overall.fscore)) 221 | final_report.append("".join(line)) 222 | 223 | for i, m in sorted(by_type.items()): 224 | line = [] 225 | line.append('%17s: ' % i) 226 | line.append('precision: %6.2f%%; ' % (100. * m.prec)) 227 | line.append('recall: %6.2f%%; ' % (100. * m.rec)) 228 | line.append('FB1: %6.2f %d\n' % (100. * m.fscore, c.t_found_guessed[i])) 229 | final_report.append("".join(line)) 230 | return final_report 231 | 232 | 233 | def end_of_chunk(prev_tag, tag, prev_type, type_): 234 | # check if a chunk ended between the previous and current word 235 | # arguments: previous and current chunk tags, previous and current types 236 | chunk_end = False 237 | 238 | if prev_tag == 'E': chunk_end = True 239 | if prev_tag == 'S': chunk_end = True 240 | 241 | if prev_tag == 'B' and tag == 'B': chunk_end = True 242 | if prev_tag == 'B' and tag == 'S': chunk_end = True 243 | if prev_tag == 'B' and tag == 'O': chunk_end = True 244 | if prev_tag == 'I' and tag == 'B': chunk_end = True 245 | if prev_tag == 'I' and tag == 'S': chunk_end = True 246 | if prev_tag == 'I' and tag == 'O': chunk_end = True 247 | 248 | if prev_tag != 'O' and prev_tag != '.' and prev_type != type_: 249 | chunk_end = True 250 | 251 | # these chunks are assumed to have length 1 252 | if prev_tag == ']': chunk_end = True 253 | if prev_tag == '[': chunk_end = True 254 | 255 | return chunk_end 256 | 257 | 258 | def start_of_chunk(prev_tag, tag, prev_type, type_): 259 | # check if a chunk started between the previous and current word 260 | # arguments: previous and current chunk tags, previous and current types 261 | chunk_start = False 262 | 263 | if tag == 'B': chunk_start = True 264 | if tag == 'S': chunk_start = True 265 | 266 | if prev_tag == 'E' and tag == 'E': chunk_start = True 267 | if prev_tag == 'E' and tag == 'I': chunk_start = True 268 | if prev_tag == 'S' and tag == 'E': chunk_start = True 269 | if prev_tag == 'S' and tag == 'I': chunk_start = True 270 | if prev_tag == 'O' and tag == 'E': chunk_start = True 271 | if prev_tag == 'O' and tag == 'I': chunk_start = True 272 | 273 | if tag != 'O' and tag != '.' and prev_type != type_: 274 | chunk_start = True 275 | 276 | # these chunks are assumed to have length 1 277 | if tag == '[': chunk_start = True 278 | if tag == ']': chunk_start = True 279 | 280 | return chunk_start 281 | 282 | 283 | def return_report(input_file): 284 | with codecs.open(input_file, "r", "utf8") as f: 285 | counts = evaluate(f) 286 | return report_notprint(counts) 287 | 288 | 289 | def main(argv): 290 | args = parse_args(argv[1:]) 291 | 292 | if args.file is None: 293 | counts = evaluate(sys.stdin, args) 294 | else: 295 | with open(args.file) as f: 296 | counts = evaluate(f, args) 297 | report(counts) 298 | 299 | 300 | if __name__ == '__main__': 301 | sys.exit(main(sys.argv)) 302 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | from transformers import BertForMaskedLM, BertTokenizer,AdamW, get_linear_schedule_with_warmup 2 | from processer import * 3 | import os 4 | import numpy as np 5 | import pandas as pd 6 | from logger import logger as logging 7 | from config import Config 8 | from prompt_model import Prompt_Based_NER,evaluate 9 | from torch.utils.data import DataLoader,RandomSampler,SequentialSampler 10 | from tqdm import trange,tqdm 11 | import torch 12 | import random 13 | from utils import save_pkl,load_pkl 14 | from torch.optim import Adam 15 | from torch import nn 16 | 17 | torch.manual_seed(1) 18 | torch.cuda.manual_seed(1) 19 | random.seed(1) 20 | 21 | # os.environ["CUDA_VISIBLE_DEVICES"]="0" 22 | 23 | config=Config() 24 | 25 | if not os.path.exists(config.output_path): 26 | os.makedirs(config.output_path) 27 | 28 | CURRENT_PATH=os.getcwd() 29 | TRAIN_PATH=os.path.join(CURRENT_PATH,'data/train.txt') 30 | EVAL_PATH=os.path.join(CURRENT_PATH,"data/eval.txt") 31 | TEST_PATH=os.path.join(CURRENT_PATH,"data/test.txt") 32 | 33 | use_gpu = torch.cuda.is_available() and config.use_gpu 34 | device = torch.device('cuda' if use_gpu else config.device) 35 | config.device = device 36 | n_gpu = torch.cuda.device_count() 37 | logging.info(f"available device: {device}，count_gpu: {n_gpu}") 38 | 39 | tokenizer = BertTokenizer.from_pretrained(config.model_name_or_path) 40 | logging.info(f"=================Tokenizer {config.model_name_or_path} Load Successfully=================") 41 | 42 | label_list = get_labels(config=config) 43 | config.label_list = label_list 44 | num_labels = len(label_list) 45 | logging.info(f"loading labels successful! the size is {num_labels}, label is: {','.join(list(label_list))}") 46 | 47 | label2id, id2label = get_label2id_id2label(config.output_path, label_list=label_list) 48 | logging.info("loading label2id and id2label dictionary successful!") 49 | 50 | if config.do_train: 51 | logging.info(f"=================Start Loading Train Dataset=================") 52 | train_text,train_label=dataset_format(TRAIN_PATH,separater=" ") 53 | assert(len(train_text)==len(train_label)) 54 | train_length_list=[len(i) for i in train_text] 55 | logging.info(f"Discription of Train Dataset: {pd.DataFrame(train_length_list).describe()}") 56 | 57 | logging.info(f"=================Start Creating Train Templates=================") 58 | train_dataset=create_template(train_text,train_label) 59 | 60 | logging.info(f"=================Preview 3 Examples=================") 61 | for i in range(5): 62 | print(train_dataset[i].text+"[MASK]",train_dataset[i].label) 63 | 64 | logging.info(f"=================Convert Train Examples=================") 65 | # if os.path.exists("train_dataset.pkl"): 66 | # train_dataset=load_pkl("train_dataset.pkl") 67 | # else: 68 | train_features,train_dataset=convert_examples_to_features(train_dataset,tokenizer) 69 | save_pkl(train_dataset,"train_dataset.pkl") 70 | train_dataloader=DataLoader(train_dataset,batch_size=config.train_batch_size,sampler=RandomSampler(train_dataset)) 71 | 72 | if config.do_eval: 73 | logging.info(f"=================Start Loading Eval Dataset=================") 74 | eval_text,eval_label=dataset_format(EVAL_PATH,separater=" ") 75 | assert(len(eval_text)==len(eval_label)) 76 | 77 | logging.info(f"=================Start Creating Eval Templates=================") 78 | eval_dataset=create_template(eval_text,eval_label) 79 | 80 | logging.info(f"=================Convert Eval Examples=================") 81 | # if os.path.exists("eval_dataset.pkl"): 82 | # eval_dataset=load_pkl("eval_dataset.pkl") 83 | # else: 84 | eval_features,eval_dataset=convert_examples_to_features(eval_dataset,tokenizer) 85 | save_pkl(eval_dataset,"eval_dataset.pkl") 86 | eval_dataloader=DataLoader(eval_dataset,batch_size=config.eval_batch_size,sampler=SequentialSampler(eval_dataset)) 87 | 88 | if config.do_test: 89 | logging.info(f"=================Start Loading Test Dataset=================") 90 | test_text,test_label=dataset_format(TEST_PATH,separater=" ") 91 | assert(len(test_text)==len(test_label)) 92 | 93 | logging.info(f"=================Start Creating Test Templates=================") 94 | test_dataset=create_template(test_text,test_label) 95 | 96 | logging.info(f"=================Convert Test Examples=================") 97 | if os.path.exists("test_dataset.pkl"): 98 | test_dataset=load_pkl("test_dataset.pkl") 99 | else: 100 | test_features,test_dataset=convert_examples_to_features(test_dataset,tokenizer) 101 | save_pkl(test_dataset,"test_dataset.pkl") 102 | test_dataloader=DataLoader(test_dataset,batch_size=config.eval_batch_size,sampler=SequentialSampler(test_dataset)) 103 | 104 | model = Prompt_Based_NER(config).to(device) 105 | 106 | if use_gpu and n_gpu > 1: 107 | model = torch.nn.DataParallel(model) 108 | 109 | no_decay = ['bias', 'LayerNorm.weight'] 110 | optimizer_grouped_parameters = [ 111 | {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 112 | 'weight_decay': 0.01}, 113 | {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 114 | 'weight_decay': 0.0} 115 | ] 116 | optimizer = Adam(optimizer_grouped_parameters, lr=config.learning_rate, eps=config.adam_epsilon) 117 | 118 | t_total = len(train_dataloader) // config.gradient_accumulation_steps * config.num_train_epochs 119 | scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=config.warmup_steps, 120 | num_training_steps=t_total) 121 | CrossEntropyloss=nn.CrossEntropyLoss() 122 | 123 | logging.info("loading AdamW optimizer、Warmup LinearSchedule and calculate optimizer parameter successful!") 124 | 125 | logging.info("====================== Running training ======================") 126 | logging.info( 127 | f"Num Examples: {len(train_dataset)}, Num Batch Step: {len(train_dataloader)}, " 128 | f"Num Epochs: {config.num_train_epochs}, Num scheduler steps: {t_total}") 129 | 130 | model.train() 131 | global_step, tr_loss, logging_loss, best_f1 = 0, 0.0, 0.0, 0.0 132 | for ep in trange(int(config.num_train_epochs), desc="Epoch"): 133 | logging.info(f"#######[Epoch: {ep}/{int(config.num_train_epochs)}]#######") 134 | model.train() 135 | for step, batch in enumerate(tqdm(train_dataloader, desc="DataLoader")): 136 | if step%100==0: 137 | logging.info(f"####[Step: {step}/{len(train_dataloader)}]####") 138 | 139 | batch = tuple(t.to(device) for t in batch) 140 | input_ids, token_type_ids, attention_mask, label_ids,ori_tokens,mask_index = batch 141 | 142 | outputs = model(input_ids, token_type_ids, attention_mask,mask_index) 143 | loss = CrossEntropyloss(outputs,label_ids) 144 | 145 | if use_gpu and n_gpu > 1: 146 | # mean() to average on multi-gpu. 147 | loss = loss.mean() 148 | 149 | if config.gradient_accumulation_steps > 1: 150 | loss = loss / config.gradient_accumulation_steps 151 | 152 | # 反向传播 153 | loss.backward() 154 | tr_loss += loss.item() 155 | 156 | # 优化器_模型参数的总更新次数，和上面的t_total对应 157 | if (step + 1) % config.gradient_accumulation_steps == 0: 158 | # 更新参数 159 | optimizer.step() 160 | scheduler.step() 161 | # 梯度清零 162 | model.zero_grad() 163 | global_step += 1 164 | 165 | if global_step%config.logging_steps == 0: 166 | tr_loss_avg = tr_loss / config.logging_steps 167 | tr_loss=0 168 | logging.info(f'tr_loss_avg= {tr_loss_avg}') 169 | 170 | if config.do_eval: 171 | logging.info("====================== Running Eval ======================") 172 | all_ori_tokens_eval = [token for token in list("".join(eval_text))] 173 | overall, by_type = evaluate(config, eval_dataloader, model, id2label, all_ori_tokens_eval) 174 | 175 | f1_score = overall.fscore 176 | 177 | # save the best performs model 178 | if f1_score > best_f1: 179 | logging.info(f"******** the best f1 is {f1_score}, save model !!! ********") 180 | best_f1 = f1_score 181 | # Take care of distributed/parallel training 182 | model_to_save = model.module if hasattr(model, 'module') else model 183 | 184 | torch.save(model.state_dict(),config.output_path+"/best_model.pth") 185 | 186 | net_dict = model.state_dict() 187 | best_model = torch.load(config.output_path+"/best_model.pth") 188 | for k, v in best_model.items(): 189 | name = k[7:] # remove `module.` 190 | net_dict[name] = v 191 | 192 | torch.save(net_dict,"best_model.pth") 193 | # model_to_save.save_pretrained(config.output_path) 194 | tokenizer.save_pretrained(config.output_path) 195 | 196 | 197 | # Good practice: save your training arguments together with the trained model 198 | torch.save(config, os.path.join(config.output_path, 'training_config.bin')) 199 | torch.save(model, os.path.join(config.output_path, 'ner_model.ckpt')) 200 | logging.info("training_args.bin and ner_model.ckpt save successful!") 201 | logging.info("NER Prompt model training successful!!!") 202 | logging.info(f"Best F1 is {best_f1}!") 203 | 204 | 205 | if config.do_test: 206 | 207 | all_ori_tokens_test = [token for token in list("".join(test_text))] 208 | overall, by_type = evaluate(config, test_dataloader, model, id2label, all_ori_tokens_test) 209 | 210 | logging.info("====================== Running test ======================") 211 | logging.info(f"Num Examples: {len(test_dataset)}, Batch size: {config.eval_batch_size}") 212 | f1_score = overall.fscore 213 | logging.info(f"**********Test F1 is {f1_score}") 214 | 215 | model.eval() 216 | 217 | pred_labels = [] 218 | ori_labels=[] 219 | 220 | for b_i, (input_ids, token_type_ids, attention_mask, label_ids,ori_tokens,masked_index) in enumerate(tqdm(test_dataloader, desc="Evaluating")): 221 | input_ids = input_ids.to(config.device) 222 | attention_mask = attention_mask.to(config.device) 223 | token_type_ids = token_type_ids.to(config.device) 224 | label_ids = label_ids.to(config.device) 225 | masked_index=masked_index.to(config.device) 226 | with torch.no_grad(): 227 | output = model.forward(input_ids, token_type_ids, attention_mask,masked_index) 228 | logits=torch.argmax(output,dim=-1) 229 | 230 | for l in logits: 231 | pred_labels.append([id2label[l.item()]]) 232 | 233 | for l in label_ids: 234 | ori_labels.append([id2label[l.tolist().index(1)]]) 235 | 236 | for l in logits: 237 | pred_label = [] 238 | for idx in l[1:]: 239 | pred_label.append(id2label[idx]) 240 | pred_labels.append(pred_label) 241 | 242 | assert len(pred_labels) == len(ori_tokens) == len(ori_labels) 243 | 244 | with open(os.path.join(config.output_path, "token_labels_test.txt"), "w", encoding="utf-8") as f: 245 | for ori_tokens, ori_labels, prel in zip(ori_tokens, ori_labels, pred_labels): 246 | for ot, ol, pl in zip(ori_tokens, ori_labels, prel): 247 | if ot in ["[CLS]", "[SEP]"]: 248 | continue 249 | else: 250 | f.write(f"{ot} {ol} {pl}\n") 251 | f.write("\n") 252 | 253 | 254 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | --------------------------------------------------------------------------------