Dataset found

[callumHub]

The paper mentions they get their data from here: https://citibikenyc.com/system-data.

The replication package has scripts which processes this raw data directly in NYC/PrepareData/StationBasicInfo.py.

The weather data is harder to find as the link in the paper did not lead to it.
Instead, google "2015 NYC weather dataset" to find: https://github.com/leokassio/weather-underground-data/blob/master/new-york-2015.csv

This dataset is not hourly but daily, however it can be used in WeatherCondition.py, if you adapt it.
Specifically:
On line 40, 42 and 43 change row[15] to row[2].

On line 33, change row[9] to row[-2],

wind is the same column.

Set the hourly values to the same as the day values, this dataset is daily.

Having access to the actual weather dataset used would still be more ideal, but at least now the replication package is useable.

Also will need to upgrade from tf version 1.x to tf version 2.x. An automated script on the files using tf 1.x and addition of 1 line of code fixes these errors.

[reisiiiiiiiiii]

@callumHub Thank you for your insights! I am currently trying to reproduce the model, but am having some difficulties with the model. Is it possible for you to share all the adjustments you have made? That would help me a lot!

Best regards
Leon

[callumHub]

@reisiiiiiiiiii
I am unable to get the 'showResults' package running still but you can easily make your own to show outputs.

There was more error messages than what I describe here, but they were all fixable. After finding the data, only a few hours of debugging to get the models to run and give outputs.

If you DM me, I can share the 'working' code (and preprocessed data) I have with you then you just have to install packages and run the model, but I find the replication package to be very hard to work with, difficult to understand and to extend for numerous reasons.
...
Since then, PyTorch Geometric Temporal has made it much easier to work with this kind of data.

I've almost finished re-implementing this paper in Pytorch Geometric, direct message me for this.

[reisiiiiiiiiii]

@callumHub
That would be great if you could send me the "working" code of this model! Do you have a preferred method of communication? Unfortunately, I couldn't find any other contact information on your profile that I could use to DM you.

[callumHub]

@reisiiiiiiiiii Sorry for the late reply. Email me: [email protected] for the "working code"

[Kitten0007]

Hello, could you share the "working" code of this model with me? It's also very helpful for me. Thank you very much!
Best regards
Ruoyan Song

[callumHub]

I pushed the working NYC portion of this replication package. This includes a link to the preprocessed data. https://github.com/callumHub/BikeGraphLSTMReplication