CMeKG_tools_Version_3/Trainer_ner.py at master · linzixuan45/CMeKG_tools_Version_3 · GitHub

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
from tqdm import tqdm
from torch.autograd import Variable
import torch.optim as optim
from torch.utils.data import TensorDataset
from torch.utils.data import DataLoader
import torch
from tools.utils import load_vocab, load_data, recover_label, get_ner_fmeasure

from All_models.model_ner.bert_lstm_crf import BERT_LSTM_CRF
from All_models.Optimizer import Lion
import os
import time
import glob

# tag-entity:{d:疾病 s:临床表现 b:身体 e:医疗设备 p:医疗程序 m:微生物类 k:科室 i:医学检验项目 y:药物}
l2i_dic = {"o": 0, "d-B": 1, "d-M": 2, "d-E": 3, "s-B": 4, "s-M": 5, "s-E": 6,
           "b-B": 7, "b-M": 8, "b-E": 9, "e-B": 10, "e-M": 11, "e-E": 12, "p-B": 13, "p-M": 14, "p-E": 15,
           "m-B": 16, "m-M": 17,
           "m-E": 18, "k-B": 19, "k-M": 20, "k-E": 21, "i-B": 22, "i-M": 23, "i-E": 24, "y-B": 25, "y-M": 26,
           "y-E": 27, "<pad>": 28, "<start>": 29, "<eos>": 30}

i2l_dic = {0: "o", 1: "d-B", 2: "d-M", 3: "d-E", 4: "s-B", 5: "s-M",
           6: "s-E", 7: "b-B", 8: "b-M", 9: "b-E", 10: "e-B", 11: "e-M", 12: "e-E", 13: "p-B", 14: "p-M",
           15: "p-E",
           16: "m-B", 17: "m-M", 18: "m-E", 19: "k-B", 20: "k-M", 21: "k-E",
           22: "i-B", 23: "i-M", 24: "i-E", 25: "y-B", 26: "y-M", 27: "y-E", 28: "<pad>", 29: "<start>",
           30: "<eos>"}

max_length = 450
batch_size = 2
epochs = 100
tagset_size = len(l2i_dic)
use_cuda = True
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")


class Trainer:
    def __init__(self, checkpoint_in='/root/医学实体识别_version_1/model_ner/medical_ner/model.pkl',
                 save_dir='model_ner/train/model/'):
        self.checkpoint = checkpoint_in
        self.medical_bert = 'checkpoint/Bert_embedding'
        self.save_model_dir = save_dir
        self.max_length = 450
        self.batch_size = 2
        self.epochs = 100
        self.tagset_size = len(l2i_dic)
        self.use_cuda = True
        self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
        """--------------------------"""
        self.init_dataset()
        self.init_model()
        self.train()

    def log(self, msg):
        msg = " {}:     {}".format(time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()), msg)
        print(msg)

    def init_model(self):
        self.model = BERT_LSTM_CRF(self.medical_bert, self.tagset_size, 768, 200, 2,
                                   dropout_ratio=0.5, dropout1=0.5, use_cuda=self.use_cuda)
        if self.checkpoint:
            self.log(f'Restoring from checkpoint: {self.checkpoint}')
            self.log(self.model.load_state_dict(
                torch.load(self.checkpoint, map_location={'cuda:0': 'cpu'}), False))
        if use_cuda:
            self.model = self.model.to(self.device)
        self.optimizer = Lion(self.model.parameters(), lr=0.0001, weight_decay=0.00001)

    def init_dataset(self):

        """step_1  all data"""
        # train_file = glob.glob('dataset/data_train_test_dev/train_*')
        # test_file = glob.glob('dataset/data_train_test_dev/test_*')
        # dev_file = glob.glob('dataset/data_train_test_dev/dev_*')
        # """step_2  finturning"""
        # train_file = glob.glob('dataset/data_finturning/train_*')
        # test_file = glob.glob('dataset/data_finturning/test_*')
        # dev_file = glob.glob('dataset/data_finturning/dev_*')
        # """step2  finturning  Row  0.685"""
        # train_file = glob.glob('dataset/data_finturning/train_Row*')
        # test_file = glob.glob('dataset/data_finturning/test_Row*')
        # dev_file = glob.glob('dataset/data_finturning/dev_Row*')

        train_Row = glob.glob('dataset/data_train_test_dev/train_Row*')
        test_Row = glob.glob('dataset/data_train_test_dev/test_Row*')
        dev_Row = glob.glob('dataset/data_train_test_dev/dev_Row*')

        train_CMeEE = glob.glob('dataset/data_train_test_dev/train_CMeEE*')
        test_CMeEE = glob.glob('dataset/data_train_test_dev/test_CMeEE*')
        dev_CMeEE = glob.glob('dataset/data_train_test_dev/dev_CMeEE*')

        train_CMedCausal = glob.glob('dataset/data_train_test_dev/train_CMedCausal*')
        test_CMedCausal = glob.glob('dataset/data_train_test_dev/train_CMedCausal*')
        dev_CMedCausal = glob.glob('dataset/data_train_test_dev/train_CMedCausal*')
        # """Row CMeEE CMedCausal"""
        # train_file = train_Row+train_CMeEE+train_CMedCausal
        # test_file = test_Row+test_CMeEE+test_CMedCausal
        # dev_file = dev_Row+dev_CMeEE+dev_CMedCausal
        """ CMeEE CMedCausal """
        train_file = train_CMeEE + train_CMedCausal
        test_file = test_CMeEE + test_CMedCausal
        dev_file = dev_CMeEE + dev_CMedCausal
        """Row  0.685"""

        vocab_file = 'checkpoint/Bert_embeading/vocab.txt'
        vocab = load_vocab(vocab_file)
        self.vocab_reverse = {v: k for k, v in vocab.items()}
        print('max_length', self.max_length)
        train_data = load_data(train_file, max_length=self.max_length, label_dic=l2i_dic, vocab=vocab, glob_dir_ls=True)
        train_ids = torch.LongTensor([temp.input_id for temp in train_data])
        train_masks = torch.LongTensor([temp.input_mask for temp in train_data])
        train_tags = torch.LongTensor([temp.label_id for temp in train_data])
        train_lenghts = torch.LongTensor([temp.lenght for temp in train_data])
        train_dataset = TensorDataset(train_ids, train_masks, train_tags, train_lenghts)
        self.train_loader = DataLoader(train_dataset, shuffle=False, batch_size=self.batch_size)
        dev_data = load_data(dev_file, max_length=self.max_length, label_dic=l2i_dic, vocab=vocab, glob_dir_ls=True)
        dev_ids = torch.LongTensor([temp.input_id for temp in dev_data])
        dev_masks = torch.LongTensor([temp.input_mask for temp in dev_data])
        dev_tags = torch.LongTensor([temp.label_id for temp in dev_data])
        dev_lenghts = torch.LongTensor([temp.lenght for temp in dev_data])
        dev_dataset = TensorDataset(dev_ids, dev_masks, dev_tags, dev_lenghts)
        self.dev_loader = DataLoader(dev_dataset, shuffle=False, batch_size=self.batch_size)
        test_data = load_data(test_file, max_length=self.max_length, label_dic=l2i_dic, vocab=vocab, glob_dir_ls=True)
        test_ids = torch.LongTensor([temp.input_id for temp in test_data])
        test_masks = torch.LongTensor([temp.input_mask for temp in test_data])
        test_tags = torch.LongTensor([temp.label_id for temp in test_data])
        test_lenghts = torch.LongTensor([temp.lenght for temp in test_data])
        test_dataset = TensorDataset(test_ids, test_masks, test_tags, test_lenghts)
        self.test_loader = DataLoader(test_dataset, shuffle=False, batch_size=self.batch_size)

    def evaluate(self, flag='dev'):
        self.model.eval()
        pred = []
        gold = []
        print('evaluate')
        if flag == 'dev':
            loader = self.dev_loader
            print(" this is dev evaluate")
        elif flag == 'test':
            loader = self.test_loader
            print(" this is test evaluate")
        with torch.no_grad():
            for i, dev_batch in enumerate(tqdm(loader)):
                sentence, masks, tags, lengths = dev_batch
                sentence, masks, tags, lengths = Variable(sentence), Variable(masks), Variable(tags), Variable(lengths)
                if self.use_cuda:
                    sentence = sentence.to(self.device)
                    masks = masks.to(self.device)
                    tags = tags.to(self.device)
                predict_tags = self.model(sentence, masks)
                loss = self.model.neg_log_likelihood_loss(sentence, masks, tags)
                pred.extend([t for t in predict_tags.tolist()])
                gold.extend([t for t in tags.tolist()])
            pred_label, gold_label = recover_label(pred, gold, l2i_dic, i2l_dic)
            print('dev loss {}'.format(loss.item()))
            pred_label_1 = [t[1:] for t in pred_label]
            gold_label_1 = [t[1:] for t in gold_label]
            acc, p, r, f = get_ner_fmeasure(gold_label_1, pred_label_1)
            print('acc:{} p: {}，r: {}, f: {}'.format(acc, p, r, f))
            return p, r, f

    def train(self):
        best_f = -100
        for epoch in range(self.epochs):
            print('epoch: {}，train'.format(epoch))
            for i, train_batch in enumerate(tqdm(self.train_loader)):
                sentence, masks, tags, lengths = train_batch
                sentence, masks, tags, lengths = Variable(sentence), Variable(masks), Variable(tags), Variable(lengths)
                if self.use_cuda:
                    sentence = sentence.to(self.device)
                    masks = masks.to(self.device)
                    tags = tags.to(self.device)
                self.model.train()
                self.optimizer.zero_grad()
                loss = self.model.neg_log_likelihood_loss(sentence, masks, tags)
                loss.backward()
                self.optimizer.step()
            print('epoch: {}，train loss: {}'.format(epoch, loss.item()))
            self.evaluate(flag='dev')
            p, r, f = self.evaluate(flag='test')
            if f > best_f:
                print('参数保存开始保存')
                best_f = f
                model_name = self.save_model_dir + "/" + 'new' + str(float('%.3f' % best_f)) + ".pkl"
                torch.save(self.model.state_dict(), model_name)
                print('保存成功')


if __name__ == "__main__":
    # checkpoint = "/root/医学实体识别_version_1/checkpoint/CMeEE/new0.914.pkl"
    # # checkpoint = '/root/医学实体识别_version_1/checkpoint/CMedCausal/new0.913.pkl'
    # # checkpoint = '/root/医学实体识别_version_1/checkpoint/CMeEE/new0.909.pkl'
    # checkpoint = '/root/医学实体识别_version_1/model_ner/medical_ner/model.pkl'
    # checkpoint = '/root/医学实体识别_version_1/checkpoint/ALL/new0.713.pkl'
    checkpoint = 'checkpoint/medical_ner/model_old.pkl'
    save_dir = "checkpoint/CMeEE_CMedCausal"
    os.makedirs(save_dir, exist_ok=True)
    train_ner = Trainer(checkpoint, save_dir)