第五课 CNN文本识别

第五课 CNN文本识别

构造输入数据

(batch_size,max_len,features_num)

卷积核大小一般设为(n*features_num)，相当于每次选取n个词，使用不同数量，不同种类的卷积核总共得到$n_{total}$个特征图，为了解决不同种类的卷积核得到不同大小的特征图，会对特征图进行一次pooling操作

网络模型架构

class Model(nn.Module):
    def __init__(self, config):
        super(Model, self).__init__()
        if config.embedding_pretrained is not None:
            self.embedding = nn.Embedding.from_pretrained(config.embedding_pretrained, freeze=False)
        else:
            self.embedding = nn.Embedding(config.n_vocab, config.embed, padding_idx=config.n_vocab - 1)
        self.convs = nn.ModuleList(
            [nn.Conv2d(1, config.num_filters, (k, config.embed)) for k in config.filter_sizes])
        # in_channels=1,out_channels=num_filters
        self.dropout = nn.Dropout(config.dropout)
        self.fc = nn.Linear(config.num_filters * len(config.filter_sizes), config.num_classes)

    def conv_and_pool(self, x, conv):
        x = F.relu(conv(x)).squeeze(3)             #[128,256（卷积核数量）,31（例，2*300大小的卷积核）]
        x = F.max_pool1d(x, x.size(2)).squeeze(2)  #[128,256]
        return x

    def forward(self, x):
        #print (x[0].shape)
        out = self.embedding(x[0])      #[128,32,300]
        out = out.unsqueeze(1)          #[128,1（颜色通道）,32,300]
        out = torch.cat([self.conv_and_pool(out, conv) for conv in self.convs], 1)  #[128,768]
        out = self.dropout(out)         #[128,768]
        out = self.fc(out)              #[128,10]
        return out