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

有谁复现出了PeMS上的结果吗? #30

Open
flowingrain opened this issue Jun 12, 2021 · 7 comments
Open

有谁复现出了PeMS上的结果吗? #30

flowingrain opened this issue Jun 12, 2021 · 7 comments

Comments

@flowingrain
Copy link

flowingrain commented Jun 12, 2021

我在TensorFlow2上兼容模式跑的,还把patience调成了20,测试集平均MAE为1.66,与报告的水平有差距

testing time: 36.1s
MAE RMSE MAPE
train 1.32 2.87 2.78%
val 1.59 3.72 3.61%
test 1.66 3.82 3.74%
performance in each prediction step
step: 01 0.99 1.88 1.96%
step: 02 1.21 2.47 2.50%
step: 03 1.38 2.97 2.93%
step: 04 1.52 3.35 3.30%
step: 05 1.62 3.65 3.61%
step: 06 1.71 3.90 3.86%
step: 07 1.78 4.09 4.08%
step: 08 1.85 4.25 4.26%
step: 09 1.90 4.38 4.42%
step: 10 1.95 4.49 4.55%
step: 11 1.99 4.59 4.67%
step: 12 2.03 4.67 4.78%
average: 1.66 3.72 3.74%
total time: 3.4min

@xingxuan2018
Copy link

请问代码data里的 'GMAN(PeMS)'这个文件在哪里能找的到?

@BitterSweets1
Copy link

我在pytorch版本上跑的,和你的结果差不多,不知道为何复现不出论文里的结果。

@wengwenchao123
Copy link

看了代码之后,发现代码里面空间注意力那边没有用分组attention,时间注意力那边用了一个mask-attention,大概是这两个变化导致我们没法复现

@wengwenchao123
Copy link

抱歉有点说错了,mask-attention是有其他用处的,那这样看就只有分组的区别了

@cuizy1017
Copy link

tensorflow的结果应该接近一些,15min看的是step03的数据,不是平均数据

@nicocccccchou
Copy link

nicocccccchou commented Jul 27, 2022

我没用文章里面说的的分组去降低时间和内存消耗,使用pytorch进行复现,除了15分钟和30分钟相对比较接近,大概只差0.02-0.04,在60分钟那个结果上差距比较大
image

@zouguojian
Copy link

复线啥啊,根本复线不出来,尤其是DCRNN更是离谱了在metr-la上

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

7 participants