文本分类论文及pytorch版复现（三）：VDCNN

最新推荐文章于 2022-12-19 19:08:22 发布

Young Panda

最新推荐文章于 2022-12-19 19:08:22 发布

阅读量679

点赞数

分类专栏： pytorch 文本分类

本文链接：https://blog.youkuaiyun.com/qq_28969139/article/details/103646076

版权

Very Deep Convolutional Networks for Text Classification

1、模型

2、代码

import torch
import torch.nn.functional as F
from torch import nn


# char-level
# embedding_dim=16, SGD, mini-batch=128, init_lr=0.01, momentum=0.9
# init_conv=He et al.,2015, use temporal batch norm without dropout.
# 29 conv layers is best.
# MaxPooling is better than KMaxPooling and Conv

class VDCNN(nn.Module):

    def __init__(self):
        super(VDCNN, self).__init__()
        num_embeddings = 5031 + 1