Chinese-Text-Classification icon indicating copy to clipboard operation
Chinese-Text-Classification copied to clipboard

Chinese-Text-Classification Project including bert-classification, textCNN and so on.

Results 3 Chinese-Text-Classification issues
Sort by recently updated
recently updated
newest added

TextCNN的代码实现 池化层的stride可以不写的, 因为只会扫描一次. stride直接设置成默认值也可以. H_out = ([seq_len-kernel_size+1] - [seq_len-kernel_size+1])/)[seq_len-kernel_size+1) +1 = 1 分子是零, stride不重要了

请问,冻结部分层的话,adam优化器那里不需要修改吗?好像需要Adam(filter(lambda p: p.requires_grad, model.parameters()))

您好!感谢作者的贡献开源。请问一下bert_classification和fine_tune bert better的训练代码,好像都没有用上validation验证集?如果要验证的话,应该怎么添加代码呢?谢谢