HiTrans/main.py at main · ljynlp/HiTrans · 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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
import argparse
import time
import utils
import dataset
import torch
import torch.autograd
import torch.nn as nn
import torch.optim as optim
from models.HiTrans import HiTrans
from models.Loss import MultiTaskLoss
from sklearn.metrics import f1_score
from torch.utils.data import DataLoader
import random
import numpy as np


class Trainer(object):
    def __init__(self, model):
        self.model = model.cuda()

        self.emo_criterion = nn.CrossEntropyLoss()
        self.spk_criterion = nn.CrossEntropyLoss()

        self.multi_loss = MultiTaskLoss(2).cuda()
        bert_params = set(self.model.bert.parameters())
        other_params = list(set(self.model.parameters()) - bert_params)
        no_decay = ['bias', 'LayerNorm.weight']
        params = [
            {'params': [p for n, p in model.bert.named_parameters() if not any(nd in n for nd in no_decay)],
             'lr': args.bert_lr,
             'weight_decay': args.weight_decay},
            {'params': [p for n, p in model.bert.named_parameters() if any(nd in n for nd in no_decay)],
             'lr': args.bert_lr,
             'weight_decay': 0.0},
            {'params': other_params,
             'lr': args.lr,
             'weight_decay': args.weight_decay},
            {"params": self.multi_loss.parameters(),
             'lr': args.lr,
             "weight_decay": args.weight_decay}
        ]

        self.optimizer = optim.Adam(params, lr=args.lr, weight_decay=args.weight_decay)
        self.scheduler = optim.lr_scheduler.ExponentialLR(self.optimizer, args.alpha)

    def train(self, data_loader):
        self.model.train()
        loss_array = []
        emo_gold_array = []
        emo_pred_array = []
        for dia_input, emo_label, spk_label, cls_index, emo_mask, spk_mask in data_loader:
            dia_input = dia_input.cuda()
            emo_label = emo_label.cuda()
            spk_label = spk_label.cuda()
            cls_index = cls_index.cuda()
            emo_mask = emo_mask.cuda()
            spk_mask = spk_mask.cuda()

            emo_output, spk_output = self.model(dia_input, cls_index, emo_mask)

            emo_output = emo_output[emo_mask]
            emo_label = emo_label[emo_mask]
            emo_loss = self.emo_criterion(emo_output, emo_label)

            spk_output = spk_output[spk_mask]
            spk_label = spk_label[spk_mask]
            spk_loss = self.spk_criterion(spk_output, spk_label)

            loss = self.multi_loss(emo_loss, spk_loss)

            loss.backward()
            torch.nn.utils.clip_grad_norm_(self.model.parameters(), args.max_grad_norm)

            self.optimizer.step()
            self.model.zero_grad()

            loss_array.append(loss.item())

            emo_pred = torch.argmax(emo_output, -1)
            emo_pred_array.append(emo_pred.cpu().numpy())
            emo_gold_array.append(emo_label.cpu().numpy())

        self.scheduler.step()
        emo_gold_array = np.concatenate(emo_gold_array)
        emo_pred_array = np.concatenate(emo_pred_array)

        f1 = f1_score(emo_gold_array, emo_pred_array, average='weighted')

        loss = np.mean(loss_array)

        return loss, f1

    def eval(self, data_loader):
        self.model.eval()
        loss_array = []
        emo_gold_array = []
        emo_pred_array = []
        with torch.no_grad():
            for dia_input, emo_label, spk_label, cls_index, emo_mask, spk_mask in data_loader:
                dia_input = dia_input.cuda()
                emo_label = emo_label.cuda()
                spk_label = spk_label.cuda()
                cls_index = cls_index.cuda()
                emo_mask = emo_mask.cuda()
                spk_mask = spk_mask.cuda()

                emo_output, spk_output = self.model(dia_input, cls_index, emo_mask)

                emo_output = emo_output[emo_mask]
                emo_label = emo_label[emo_mask]
                emo_loss = self.emo_criterion(emo_output, emo_label)

                spk_output = spk_output[spk_mask]
                spk_label = spk_label[spk_mask]
                spk_loss = self.spk_criterion(spk_output, spk_label)

                loss = self.multi_loss(emo_loss, spk_loss)

                loss_array.append(loss.item())

                emo_pred = torch.argmax(emo_output, -1)
                emo_pred_array.append(emo_pred.cpu().numpy())
                emo_gold_array.append(emo_label.cpu().numpy())

        emo_gold_array = np.concatenate(emo_gold_array)
        emo_pred_array = np.concatenate(emo_pred_array)

        f1 = f1_score(emo_gold_array, emo_pred_array, average='weighted')
        loss = np.mean(loss_array)

        return loss, f1

    def save(self, path):
        torch.save(self.model.state_dict(), path)

    def load(self, path):
        self.model.load_state_dict(torch.load(path))


if __name__ == '__main__':
    parser = argparse.ArgumentParser()

    parser.add_argument('--d_model', type=int, default=768)
    parser.add_argument('--d_ff', type=int, default=768)
    parser.add_argument('--heads', type=int, default=6)
    parser.add_argument('--layers', type=int, default=1)
    parser.add_argument('--input_max_length', type=int, default=512)
    parser.add_argument('--hidden_dim', type=int, default=768)
    parser.add_argument('--dropout', type=float, default=0.5)
    parser.add_argument('--lr', type=float, default=5e-5)
    parser.add_argument('--bert_lr', type=float, default=1e-5)
    parser.add_argument('--weight_decay', type=float, default=1e-5)
    parser.add_argument('--max_grad_norm', type=float, default=1.0)
    parser.add_argument('--epochs', type=int, default=20)
    parser.add_argument('--batch_size', type=int, default=8)
    parser.add_argument('--alpha', type=float, default=0.95)
    parser.add_argument('--seed', type=int, default=123)
    parser.add_argument('--device', type=int, default=0)

    parser.add_argument('--evaluate', action='store_true')
    args = parser.parse_args()

    logger = utils.get_logger("./log/HiTrans_{}.txt".format(time.strftime("%m-%d_%H-%M-%S")))
    logger.info(args)

    torch.cuda.set_device(args.device)

    seed = args.seed
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)

    logger.info("Loading data...")

    (train_set, dev_set, test_set), vocab = dataset.load_data(args.input_max_length)
    if args.evaluate:
        dev_loader = DataLoader(dataset=dev_set, batch_size=args.batch_size, collate_fn=dataset.collate_fn)
        test_loader = DataLoader(dataset=test_set, batch_size=args.batch_size, collate_fn=dataset.collate_fn)
    else:
        train_loader = DataLoader(dataset=train_set, batch_size=args.batch_size, collate_fn=dataset.collate_fn,
                                  shuffle=True)
        dev_loader = DataLoader(dataset=dev_set, batch_size=args.batch_size, collate_fn=dataset.collate_fn)
        # test_loader = DataLoader(dataset=test_set, batch_size=args.batch_size, collate_fn=dataset.collate_fn)

    model = HiTrans(args.hidden_dim,
                    len(vocab.label2id),
                    d_model=args.d_model,
                    d_ff=args.d_ff,
                    heads=args.heads,
                    layers=args.layers,
                    dropout=args.dropout)

    trainer = Trainer(model)

    if args.evaluate:
        trainer.load("./checkpoint/model.pkl")
        dev_loss, dev_f1 = trainer.eval(dev_loader)
        logger.info("Dev Loss: {:.4f} F1: {:.4f}".format(dev_loss, dev_f1))
        test_loss, test_f1 = trainer.eval(test_loader)
        logger.info("Test Loss: {:.4f} F1: {:.4f}".format(test_loss, test_f1))
    else:
        best_f1 = 0.0
        for epoch in range(args.epochs):
            train_loss, train_f1 = trainer.train(train_loader)
            logger.info("Epoch: {} Train Loss: {:.4f} F1: {:.4f}".format(epoch, train_loss, train_f1))
            dev_loss, dev_f1 = trainer.eval(dev_loader)
            logger.info("Epoch: {} Dev Loss: {:.4f} F1: {:.4f}".format(epoch, dev_loss, dev_f1))
            # test_loss, test_f1 = trainer.eval(test_loader)
            # logger.info("Test Loss: {:.4f} F1: {:.4f}".format(test_loss, test_f1))
            logger.info("---------------------------------")
            if best_f1 < dev_f1:
                best_f1 = dev_f1
                trainer.save("./checkpoint/model.pkl")

        logger.info("Best Dev F1: {:.4f}".format(best_f1))