search

tangxianfeng / STDN

Code for our Spatiotemporal Dynamic Network
275 stars 98 forks source link

I can't find the paper name in KDD accepted paper 2018 #1

Open LeeDoYup opened 5 years ago

LeeDoYup commented 5 years ago

Is the paper STDN accepted in KDD 2018? I can't find the name in accepted paper list. Could you let me know the link?

dreampppaper commented 5 years ago

Hi~ Can you provide your code about how to generate data? I have some confusion about how to generate data. Thank you！

tangxianfeng commented 5 years ago

Hi @LeeDoYup sorry for late reply. The paper is accepted by AAAI 2019. Modeling Spatial-Temporal Dynamics for Traffic Prediction: https://arxiv.org/abs/1803.01254

@dreampppaper You can follow the detail description in the paper for data generation.

LeeDoYup commented 5 years ago

Thank you and sincerly congrats in 16% !

dreampppaper commented 5 years ago

Thank you for your quick reply！ Congratulations on your paper accepted by AAAI！ This is an outstanding work！ Please forgive my clumsiness. Although I have read your description of the data many times in your paper, I still feel a bit confused when reading the code. As follows: flow_feature_curr_out = flow_data[0, real_t, x, y, :, :] flow_feature_curr_in = flow_data[0, real_t, :, :, x, y] flow_feature_last_out_to_curr = flow_data[1, real_t - 1, x, y, :, :] flow_feature_curr_in_from_last = flow_data[1, real_t - 1, :, :, x, y] I know that the first two cows of your code are the inflow/outflow matrices in the t-th time interval you described in the paper. But I have a question about the code of the last two lines. From the perspective of the variable name you defined，it means departing/arriving from this region/other regions in last time interval to other regions/this region in real_t time interval. If it is the meaning I described above，why flow_data[1, real_t, :, :, x, y] != flow_data[0, real_t, :, :, x, y]? if not,could you tell me the meaning of last tow lines. Thank you! @tangxianfeng @LeeDoYup Do you tell me the answer about this question?

LeeDoYup commented 5 years ago

@dreampppaper Sorry for my closing this issue. I didn't see your remaining question. Also I have another question about data preprocessing. Following these codes, only the half of test data is evaluated. I think it is somewhat weird, because only 10 days are evaluated in total 20 days test duration.

tangxianfeng commented 5 years ago

That's because you need some extra space to sample data at previous days.

Leerw commented 5 years ago

@tangxianfeng Thank you for your great work! I have an understanding barrier for the data .npz you provided, like issued above,

flow_feature_curr_out = flow_data[0, real_t, x, y, :, :]
flow_feature_curr_in = flow_data[0, real_t, :, :, x, y]
flow_feature_last_out_to_curr = flow_data[1, real_t - 1, x, y, :, :]
#real_t - 1 is the time for in flow in longflow1
flow_feature_curr_in_from_last = flow_data[1, real_t - 1, :, :, x, y]

what are the meaning of these variables? what are flow_data[0,...] and flow_data[1,...] stand for? Can you explain your data format in detail?

Note: problem solved, see data format at this