Why are results in readme lower than results in paper?

[woshiyyya]

Why are results in readme lower than results in paper?

[Wsl-hfut-nwpu]

你看那个小然后调整那个超参数就行。

[shaonan1993]

这个结果对的呀，没有低.....

[Gabriellamin]

你看看论文里的结果，和这个不一样

…

---Original--- From: "邵楠Griffiths"<[email protected]&gt; Date: Mon, Sep 23, 2019 14:22 PM To: "woshiyyya/DFGN-pytorch"<[email protected]&gt;; Cc: "Gabriellamin"<[email protected]&gt;;"Author"<[email protected]&gt;; Subject: Re: [woshiyyya/DFGN-pytorch] Why are results in readme lower than results in paper? (#16) 这个结果对的呀，没有低..... — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[shaonan1993]

你看看论文里的结果，和这个不一样
…
---Original--- From: "邵楠Griffiths"<[email protected]> Date: Mon, Sep 23, 2019 14:22 PM To: "woshiyyya/DFGN-pytorch"<[email protected]>; Cc: "Gabriellamin"<[email protected]>;"Author"<[email protected]>; Subject: Re: [woshiyyya/DFGN-pytorch] Why are results in readme lower than results in paper? (#16) 这个结果对的呀，没有低..... — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

论文里是EM/F1: 33.62/59.82，你这个结果是EM/F1: 33.77/59.89。

[Gabriellamin]

论文里的answer em/f1是56.31/69.69，这里是55.67/69.38...分别低0.64/0.31

[shaonan1993]

论文里的answer em/f1是56.31/69.69，这里是55.67/69.38...分别低0.64/0.31

那你调一下lambda好了......

[woshiyyya]

hi, 论文中公布的结果是leaderboard上面test set的结果，而github上是dev集的结果，所以二者会有区别。

[Gabriellamin]

测试集上测试的流程是怎样的呢？可以公布一下代码吗