search

tech-srl / bottleneck

Code for the paper: "On the Bottleneck of Graph Neural Networks and Its Practical Implications"
MIT License
91 stars 22 forks source link

Do task have the attribute of DICTIONARY #4

Closed UnBuen closed 3 years ago

UnBuen commented 3 years ago

image

Dear tech-srl, maybe hear is a mistake.

urialon commented 3 years ago

Hi @UnBuen , Thank you for reporting this!

The task should indeed be NEIGHBORS_MATCH. This flag can also be omitted because this task is the default.

I will fix this in the README.

Best, Uri

UnBuen commented 3 years ago

Dear Uri, Sorry to bother you again~

when I run command python run-gcn-2-8.py, There maybe is a problem with the number of training examples and test examples. image

Have you encountered this problem, or do you have any solutions? Best, Unbuen

urialon commented 3 years ago

Hi @UnBuen , This is not a bother at all.

With depth=2, the number of leaves is 4, so combinatorially there is only a small number of possible different examples.

depth=2 is really a toy example.

UnBuen commented 3 years ago

Hi @UnBuen , This is not a bother at all.

With depth=2, the number of leaves is 4, so combinatorially there is only a small number of possible different examples.

depth=2 is really a toy example.

Hi, Uri, Oh, I understand now, thanks a lot

UnBuen commented 3 years ago

Hi @urialon , I ran into another problem when I ran the tf-gnn-samples/run_qm9_benchs_fa.py and tf-gnn-samples/run_varmisuse_benchs_fa.py

My development(cent os 7 and python3.6) image image

Problem description image image

If you have encountered this problem, please help me or give me some suggestions, thanks. I saw other people also have this problem on github and some blog, but many people do not have a correct solution.

If you have time, please give me some suggestions, it's okay if you are busy, I can also consult my seniors and mentors.

Wish you success in your studies and work. Best, UnBuen.

urialon commented 3 years ago

Hi UnBuen, Sorry for the delayed response!

I did encounter this in the past. The run_qm9_benchs_fa.py script is originally from Microsoft's tf-gnn-samples repo. It is simply a loop that runs train.py 5 times. The problem is that if train.py fails, it hides the errors. One way to see what was the error is to check the directory that you provided to run_qm9_benchs_fa.py as the first argument. For example, if you ran:

python run_qm9_benchs_fa.py myLogsDir/

then check myLogsDir/ to see if you can identify the error. If you can't find the error, try running train.py using Microsoft's instructions here: https://github.com/tech-srl/bottleneck/tree/main/tf-gnn-samples#running.

Let me know how it goes. Best, Uri

UnBuen commented 3 years ago

Hi UnBuen, Sorry for the delayed response!

I did encounter this in the past. The run_qm9_benchs_fa.py script is originally from Microsoft's tf-gnn-samples repo. It is simply a loop that runs train.py 5 times. The problem is that if train.py fails, it hides the errors. One way to see what was the error is to check the directory that you provided to run_qm9_benchs_fa.py as the first argument. For example, if you ran:

python run_qm9_benchs_fa.py myLogsDir/

then check myLogsDir/ to see if you can identify the error. If you can't find the error, try running train.py using Microsoft's instructions here: https://github.com/tech-srl/bottleneck/tree/main/tf-gnn-samples#running.

Let me know how it goes. Best, Uri

Hi Uri, Thank you for your valuable comments. I have solved my problem. The reason for the previous failure is due to the dataset and data processing part. I have found a solution from https://github.com/microsoft/tf-gnn-samples,. Thank you again for your patience. Best, UnBuen

urialon commented 3 years ago

Great, glad to hear! Let me know if you have any further questions.