search

Tiiiger / SGC

official implementation for the paper "Simplifying Graph Convolutional Networks"
MIT License
831 stars 145 forks source link

confusion about the network structure #10

Closed hl4080 closed 5 years ago

hl4080 commented 5 years ago 
hello,I have run your code and it's dramatically faster than other GCN framework.
But could you tell me why your hidden unit is set to 0 and I try to change it,but it make no difference to the test precision.
Thanks!
Tiiiger commented 5 years ago

Hi @hl4080,

I am not sure what you are referring to by 'hidden unit is set to 0'. Can you point me to the line of code?

hl4080 commented 5 years ago

@Tiiiger Thanks,Tiiiger In the arg.py,you wrote parser.add_argument('--hidden', type=int, default=0, help='Number of hidden units.') I have tried to chang the default number,but it makes no difference to the test precision,and it does change the time spent on training.I wonder why.

Tiiiger commented 5 years ago

This argument is meant for GCN model. You can see https://github.com/Tiiiger/SGC/blob/master/models.py#L66 that we did not use the argument when creating a SGC model.

I am surprised that you observe that it change the time spent on training. Can you please double check that you are using SGC? Maybe try running more times? it could be a variance thing.

Let me know you have further question.

Tiiiger commented 5 years ago

I am closing this if you don't have further questions.

hl4080 commented 5 years ago

Oh,thank you! That's very kind of you!