text_classify/train_eval.py at master · mathCrazyy/text_classify · 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
from data import *

#from model_embedding import Model
#from models.lstm_embedding import Model

import torch
import tqdm
from torchtext.data import Iterator, BucketIterator
import torch.nn.functional as F
from sklearn.metrics import classification_report
from sklearn import metrics

import numpy as np
from tensorboardX import SummaryWriter
import time

def evaluate(config, model, eval_iter, test=False):
    model.eval()
    val_loss = 0.0
    all_acc = 0.0
    predict_all = np.array([], dtype=int)
    labels_all = np.array([], dtype=int)
    loss_total=0.0
    with torch.no_grad():
        for batch in eval_iter:
            preds = model(batch.context)
            loss=F.cross_entropy(preds.cpu(),batch.label_id.cpu())
            loss_total+=loss.item()
            predic=torch.max(preds.data,1)[1].cpu().numpy()
            predict_all = np.append(predict_all, predic)
            labels_all = np.append(labels_all, batch.label_id.cpu())

        acc=metrics.accuracy_score(labels_all,predict_all)
        # class_report=classification_report(predict_all,labels_all)
        class_report = classification_report(labels_all, predict_all)
        # print(labels_all)
        # print(predict_all)
    return loss_total/len(eval_iter),acc, class_report


def test(config, model, TEXT, sentence):
    sentence_seq=[TEXT.vocab.stoi[one] for one in sentence]
    need_pad=config.sen_max_length-len(sentence_seq)
    for _ in range(need_pad):
        sentence_seq.append(1)

    example=torch.Tensor(sentence_seq).long().to(config.device)
    example=example.unsqueeze(1)

    preds=model(example)
    predic=torch.max(preds.data,1)[1].cpu().numpy()
    return predic


def train(config, model, train_iter, valid_iter, test_iter):
    model.train()
    optimizer = torch.optim.Adam(model.parameters(), lr=config.learning_rate)

    eval_best_loss = float("inf")
    epochs = config.epochs
    writer=SummaryWriter(log_dir=config.log_path+"/"+time.strftime("%m-%d_%H.%M",time.localtime()))
    total_batch=0
    last_improve=0
    flag=False

    for epoch in range(1, epochs + 1):
        # if epoch % 5 ==0:
        #    for p in optimizer.param_groups:
        #        p["lr"]*=0.9
        running_loss = 0.0
        runing_corrects = 0
        model.train()

        for batch in train_iter:
            total_batch+=1
            model.zero_grad()
            preds = model(batch.context)
            ## 应当是10行，5列的样子
            # y_p = batch.label_id.squeeze(1)
            y_p = batch.label_id
            # https://blog.csdn.net/ccbrid/article/details/90610599
            loss = F.cross_entropy(preds, y_p.long()).to(config.device)

            loss.backward()
            optimizer.step()
            if(total_batch%100==0):
                pred_res = torch.max(preds.data, 1)[1].cpu()
                train_acc = metrics.accuracy_score(y_p.cpu(), pred_res)
                eval_loss, eval_acc, eval_report = evaluate(config, model, valid_iter)
                test_loss, test_acc, test_report = evaluate(config, model, test_iter)
                print("train_loss: ",loss,"train_acc: ",train_acc,total_batch)
                print("eval_loss: ",eval_loss,"eval_acc: ",eval_acc,total_batch)
                print("test_loss: ",test_loss,"test_acc: ",test_acc,total_batch)
                if eval_loss < eval_best_loss:
                    eval_best_loss = eval_loss
                    torch.save(model.state_dict(), config.save_path)
                    last_improve=total_batch

                writer.add_scalar("loss/train",loss.item(),total_batch)
                writer.add_scalar("loss/dev",eval_loss,total_batch)
                writer.add_scalar("acc/train",train_acc,total_batch)
                writer.add_scalar("acc/dev",eval_acc,total_batch)
                model.train()
            if total_batch-last_improve>config.require_improvement and last_improve!=0:
                print(total_batch-last_improve)
                print(config.require_improvement)
                print("超过",config.require_improvement,"轮次没有提升并退出")
                print("eval_report: ", eval_report)
                print("test_report", test_report)
                flag=True
                break
        if flag:
            break


"""

config = Config()

nh = 64
model = SimpleLSTMBaseline(TEXT, nh).to(device)

train(config, model, train_iter, valid_iter, test_iter)


"""