search

snap-stanford / relbench

RelBench: Relational Deep Learning Benchmark
https://relbench.stanford.edu
MIT License
223 stars 41 forks source link

Error when running script #237

Closed Yuuuuuuu1328 closed 4 months ago

Yuuuuuuu1328 commented 4 months ago

Dear relbench team,

python gnn_node.py --dataset rel-f1 --task driver-position

Traceback (most recent call last): File "gnn_node.py", line 70, in <module> data, col_stats_dict = make_pkey_fkey_graph( File "\relbench\modeling\graph.py", line 83, in make_pkey_fkey_graph to_unix_time(table.df[table.time_col]) File "\relbench\modeling\utils.py", line 14, in to_unix_time unix_time = ser.astype(int).values

raise TypeError( TypeError: Converting from datetime64[ns] to int32 is not supported. Do obj.astype('int64').astype(dtype) instead

Yuuuuuuu1328 commented 4 months ago

unix_time = ser.astype('int64').values works

I've been working closely with your RelBench framework, specifically version 1.0.0, and I've noticed an area where a slight enhancement could potentially improve the user experience.

There seems to be a detail that requires attention, and I believe addressing it would make the framework even more robust and user-friendly. If possible, I'd recommend considering this refinement in an upcoming update.

Additionally, I've observed that when using versions other than 1.0.0, there are minor issues that crop up, which might be related to compatibility or changes in the API. To help prevent confusion for future users, I suggest updating the README file to clearly indicate any known compatibility constraints or recommended version ranges. This could include specifying that certain features or bug fixes are exclusive to version 1.0.0 or later.

By keeping the documentation up-to-date with such information, new adopters of RelBench can have a smoother onboarding process and avoid potential pitfalls early on in their projects.

Thank you for developing such a valuable tool for the community. Your work is greatly appreciated, and I'm looking forward to seeing how RelBench evolves in the future.

yiweny commented 4 months ago

Hi @Yuuuuuuu1328 I cannot reproduce your problem. What are your package versions? Mine is the following

Pandas version: 1.5.2, NumPy version: 1.26.0
Yuuuuuuu1328 commented 4 months ago

Hi @Yuuuuuuu1328 I cannot reproduce your problem. What are your package versions? Mine is the following

Pandas version: 1.5.2, NumPy version: 1.26.0

Pandas version: 2.0.3 , NumPy version: 1.23.5

Thank you very much for your reply. Here's my version.

yiweny commented 4 months ago

Can you degrade the pandas version and see if the code works by default?

rishabh-ranjan commented 4 months ago

Pandas version: 2.0.3 and Numpy 1.24.4 works on my end