MengzhangLI/STFGNN

Questions about the dimensions of datasets

Opened this issue · 6 comments

QEASONT commented

I am trying to use Temporal_Graph_gen.py to process METR-LA.
6357191CBE7F0D7BFD52581B95012938
README said that in PEMS08 dataset, the shape is (Total_Time_Steps, Node_Number), however, when I load the PEMS08, the shape is (17856,170,3), Could you please tell me what does the third dimension means?
{FF0358DF-37BC-04CE-50DC-065F0181F55B}

MengzhangLI commented

Hi,

17856 is time step points, 170 is number of roads/ sensors and 3 is number of metric such as speed, volume and so on.

In traffic forecasting paper, many paper use more than one metric for forecasting, but they usually use 1 metric (i.e., speed) because of current widely used dataset such as METR-LA and PEMS-BAY only have information of speed.

You can checkout the dimension 3 and get the speed value of it. If I remember well, it is in [:, :, 0], you could check it out.

Best,

1834399017 commented

我想请问一下，这是在Linus上跑的这部分嘛

MengzhangLI commented

我想请问一下，这是在Linus上跑的这部分嘛

嗯是的，用的学校的服务器。

1834399017 commented

您好，我想一下，您在代码中设置的间隔时间是多久的呢？在哪里可以改变一下？还有在代码中似乎没有找到门控部分的代码，我在pems08中训练后与论文中的效果改进有点大，可能是什么原因呢？（因为没有服务器可用，03，04，07跑不起）希望您能帮我解决一下，万分感谢！

MengzhangLI commented

时间间隔等所有配置能改的都在config json里 https://github.com/MengzhangLI/STFGNN/blob/master/config/PEMS08/individual_3layer_12T.json

门空代码在这里:

STFGNN/models/stsgcn_4n_res.py

Line 258 in a02feee

data_time_axis = data_left * data_right

PEMS08结果差的原因我觉得不是我的代码问题，我在PEMS数据集的结果都是用这个repo的代码并且跑好了很多次.

1834399017 commented

非常感谢您的回答！请问可以加一下您的联系方式嘛？我有一些问题想和您探讨一下，我得q1834399017，期待您的回复