Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Beam Search finalize() bug: not update parents in the "body" #2

Open
AaronYALai opened this issue May 8, 2018 · 5 comments
Open

Beam Search finalize() bug: not update parents in the "body" #2

AaronYALai opened this issue May 8, 2018 · 5 comments

Comments

@AaronYALai
Copy link

return (time + 1), outputs_ta, parents

The "body" function for the tf.while_loop extracts final decoding results time step by time step.

But the state "parents" has not been updated in the body function!

def body(time, outputs_ta, parents): 
    ... (no update of parents) ...
    return (time + 1), outputs_ta, parents

This should be as the following:

return (time + 1), outputs_ta, input_t.parents

since parents for the next step are stored in "input_t" which is extracted for the current time step.

@interxuxing
Copy link

@AaronYALai maybe you are right, i modified the beaming search decoding according to your comment, the performance has 5% improvement on my own dataset.

@JunweiLiang
Copy link

@AaronYALai @interxuxing I think it should the next parents should be gather_helper(input_t.parents, parents) as the parent_idx is the traceback to the last timestep's parent at each timestep.

@kim-yhow
Copy link

kim-yhow commented Nov 7, 2019

@AaronYALai @interxuxing I think it should the next parents should be gather_helper(input_t.parents, parents) as the parent_idx is the traceback to the last timestep's parent at each timestep.

你好,你最近也在也就公式图片到latex么,我跑了这个代码,发现EM指标只有51%,没有达到论文中的76%,这份代码是不是与论文的里提到的方法不一样。我没有看到本篇文章使用了Row encoder 的方法

@pageedward
Copy link

pageedward commented Jun 18, 2021

@AaronYALai @interxuxing I think it should the next parents should be gather_helper(input_t.parents, parents) as the parent_idx is the traceback to the last timestep's parent at each timestep.

你好,你最近也在也就公式图片到latex么,我跑了这个代码,发现EM指标只有51%,没有达到论文中的76%,这份代码是不是与论文的里提到的方法不一样。我没有看到本篇文章使用了Row encoder 的方法

你的EM51是跑了迭代次数多少 ,衰减多少,我按照默认配置跑6+13(decay),em只有36.57+%

@pageedward
Copy link

@AaronYALai @interxuxing I think it should the next parents should be gather_helper(input_t.parents, parents) as the parent_idx is the traceback to the last timestep's parent at each timestep.

你好,你最近也在也就公式图片到latex么,我跑了这个代码,发现EM指标只有51%,没有达到论文中的76%,这份代码是不是与论文的里提到的方法不一样。我没有看到本篇文章使用了Row encoder 的方法
你的EM51是跑了迭代次数多少 ,衰减多少,我按照默认配置跑6+13(decay),em只有36.57+%

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants