search

panzheyi / ST-MetaNet

The codes and data of paper "Urban Traffic Prediction from Spatio-Temporal Data Using Deep Meta Learning"
MIT License
194 stars 73 forks source link

Question regarding Seq2Seq input data #6

Open razvanc92 opened 3 years ago

razvanc92 commented 3 years ago

Hello, I'm having a question reading the input of Seq2Seq model. In the paper you mentioned that "The features of nodes are added as the additional inputs." From my understanding on the last dimension the input should have a size of: 9 (node features) + 8 (neighbors) * 2 (x,y coordinates). I searched everywhere in the MyGRUCell and I can not find such shape, furthermore I can see that the features are not used in the cell.

Thank you in advance and looking forward to hearing from you.

panzheyi commented 3 years ago

Hi, The seq2seq.py file refers to the seq2seq architectures in ST-MetaNet.

The codes of original baselines (including Seq2Seq) are not included because they tackle features in a different way.

For simplicity, this repository includes the major codes of ST-MetaNet and a variant of Seq2Seq, to present the improvement.

If you want to reproduce our original baselines, you can directly modify dataloader.py to concatenate static features with the dynamic temporal readings (all static features and dynamic readings are provided in dataloader.py).