准确率问题

[dinghan007]

您好，我用您的代码跑了这几个数据集，发现实现结果跟Paper上的低了1%或者2%。请问是什么问题？感谢回复

[prasys]

你好！

我们觉得z喝个是‘normal。这个script 没有’fixed random number' （seed value)
对不起，如果我的华语不好

谢谢你

[dinghan007]

Sorry，I can't understand the Chinese you written. So can you write english answer. Thanks.------------------ 原始邮件 ------------------ 发件人: "Pradeesh"<notifications@github.com> 发送时间: 2019年12月5日(星期四) 上午7:07 收件人: "AlexYangLi/ABSA_Keras"<ABSA_Keras@noreply.github.com>; 抄送: "dinghan007"<490213124@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [AlexYangLi/ABSA_Keras] 准确率问题 (#6) 我们觉得z喝个是‘normal。这个script 没有’fixed random number' （seed value) 对不起，如果我的华语不好 谢谢你 &mdash; You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[prasys]

it's basically because neural network (as you know are non-deterministic) , in the paper or the in code - we do not know the magical number or the seed number. So even if I run the code again , I am getting a slightly different results

So if it's within some % ,. Hope this helps