libcity / bigscity-libcity-datasets Goto Github PK

File “/Bigscity-LibCity-master/libcity/model/traffic_speed_prediction/STMGAT.py", line 3, in
from dgl.nn.pytorch.conv import GATConv
ModuleNotFoundError: No module named 'dgl.nn.pytorch'

I have one question regarding the config files of PEMSD4 and PEMSD8:

Why the "set_weight_link_or_dist" is "link" and "init_weight_inf_or_zero" is "zero"?
Should they be "dist" and "inf", just like the setting of "PEMSD7(M)"?

Another similar question on Beijing_Subway:

Thank you!

Hi. Thanks for such a amazing repo. I intend to construct a dataset from raw data. For example, constructing grid-based demand dataset for traffic demand prediction task from NYCBike and NYCTaxi data. Unfortunately, I do not find any related script can do such construction. The most related script is nyctaxi_od.py. It seems does not produce grid-based data?

Would you like to give me some suggestions?

Whenever I successfully train a model and then evaluate it. The value produced for MAPE is inf. Every other value, like MAE, RMSE etc shows correctly. But MAPE shows inf value. What's causing this issue? I am unable to figure it out. Any help would be much appreciated.

The gridod file is needed when running the STDN (Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction), but is missing in the Google Drive.

I created it following your code "nycbike20160708.py". You can now access it via the following link:

https://drive.google.com/file/d/1KOIj8w2Nf9Sd5TDV7IuJwZOUxOz3PJ6_/view?usp=sharing

Hi,

May I know how did you calculate the statistics for user numbers? They are different from the original paper which published the dataset you used.

link: https://www.citibikenyc.com/system-data
BikeNYC数据，202101-citibike-tripdata.csv 及以前的数据字段为：'tripduration, starttime, stoptime, start station id, start station name, start station latitude, start station longitude, end station id, end station name, end station latitude, end station longitude, bikeid, usertype, birth year, gender' 一共15个，而之后的'ride_id, rideable_type, started_at, ended_at, start_station_name, start_station_id, end_station_name, end_station_id, start_lat, start_lng, end_lat, end_lng, member_casual' 只有13个，当前数据预处理脚本nycbike.py 只能处理之前的数据。对于之后的数据无法处理。

Dear Authors,

I want to add one NN layer to DCRNN, STGCN, MTGNN before the final output. Do I just change the file in traffic_speed_predition folder?

I am just not very sure which method should I change. Do I rewrite the forward method in the class in each file?

一开始完全按照指引操作出现报错，似乎是显卡算力问题：

于是尝试安装新版cuda:
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
由于requirements.txt中的torch版本较低，所以我删去了那一行，使用的是安装新cuda对应的torch版本；但是后续尝试了很多次，都会出现相同的错误：即在“encoding trajectory”阶段出现“已杀死”
PS：内存64g没有出现内存不够的情况，且没有他人进行kill

Hi, thanks for the incredible work. It seems the expected field was missing during the transform process.

Describe the bug

The field 'dyna_id' in the result file xx.dyna generated by foursquare.py was None.
ref. code:

Code to Reproduce

https://github.com/LibCity/Bigscity-LibCity-Datasets/blob/145f66256a9cc056826abe06d33d8d0e13c926cf/foursquare.py

Expected behavior

dyna_id should be clarified in dyna['dyna_id'] = dyna.index or other ways.

Actual result

output file (sample):

dyna_id,type,time,entity_id,location
,trajectory,2012-04-03T14:00:09Z,0,2388
,trajectory,2012-04-04T14:00:09Z,0,2388
...

Versions

Windows-10-10.0.19043-SP0
Python 3.8.12 (default, Oct 12 2021, 03:01:40) [MSC v.1916 64 bit (AMD64)]