sundea/text-classification

1# coding: UTF-82import torch3import torch.nn as nn4import torch.nn.functional as F5import numpy as np6 7 8class Config(object):9 10    """配置参数"""11    def __init__(self, dataset, embedding):12        self.model_name = 'FastText'13        self.train_path = dataset + '/data/train.txt'                                # 训练集14        self.dev_path = dataset + '/data/dev.txt'                                    # 验证集15        self.test_path = dataset + '/data/test.txt'                                  # 测试集16        self.class_list = [x.strip() for x in open(17            dataset + '/data/class.txt', encoding='utf-8').readlines()]              # 类别名单18        self.vocab_path = dataset + '/data/vocab.pkl'                                # 词表19        self.save_path = dataset + '/saved_dict/' + self.model_name + '.ckpt'        # 模型训练结果20        self.log_path = dataset + '/log/' + self.model_name21        self.embedding_pretrained = torch.tensor(22            np.load(dataset + '/data/' + embedding)["embeddings"].astype('float32'))\23            if embedding != 'random' else None                                       # 预训练词向量24        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')   # 设备25 26        self.dropout = 0.5                                              # 随机失活27        self.require_improvement = 1000                                 # 若超过1000batch效果还没提升，则提前结束训练28        self.num_classes = len(self.class_list)                         # 类别数29        self.n_vocab = 0                                                # 词表大小，在运行时赋值30        self.num_epochs = 20                                            # epoch数31        self.batch_size = 128                                           # mini-batch大小32        self.pad_size = 32                                              # 每句话处理成的长度(短填长切)33        self.learning_rate = 1e-3                                       # 学习率34        self.embed = self.embedding_pretrained.size(1)\35            if self.embedding_pretrained is not None else 300           # 字向量维度36        self.hidden_size = 256                                          # 隐藏层大小37        self.n_gram_vocab = 250499                                      # ngram 词表大小38 39 40'''Bag of Tricks for Efficient Text Classification'''41 42 43class Model(nn.Module):44    def __init__(self, config):45        super(Model, self).__init__()46        if config.embedding_pretrained is not None:47            self.embedding = nn.Embedding.from_pretrained(config.embedding_pretrained, freeze=False)48        else:49            self.embedding = nn.Embedding(config.n_vocab, config.embed, padding_idx=config.n_vocab - 1)50        self.embedding_ngram2 = nn.Embedding(config.n_gram_vocab, config.embed)51        self.embedding_ngram3 = nn.Embedding(config.n_gram_vocab, config.embed)52        self.dropout = nn.Dropout(config.dropout)53        self.fc1 = nn.Linear(config.embed * 3, config.hidden_size)54        # self.dropout2 = nn.Dropout(config.dropout)55        self.fc2 = nn.Linear(config.hidden_size, config.num_classes)56 57    def forward(self, x):58 59        out_word = self.embedding(x[0])60        out_bigram = self.embedding_ngram2(x[2])61        out_trigram = self.embedding_ngram3(x[3])62        out = torch.cat((out_word, out_bigram, out_trigram), -1)63 64        out = out.mean(dim=1)65        out = self.dropout(out)66        out = self.fc1(out)67        out = F.relu(out)68        out = self.fc2(out)69        return out70