GNN-MAGNA
GNN-MAGNA copied to clipboard
xjtuwgt
About hardware information
Hello author, I'm a student who interested in your works when reading Multi-hop Attention Graph Neural Networks Can I get the information about the hardware that you train your model I cloned your repository and I on server in my lab which has Ubuntu 16.04, 32GB RAM
The code build well but it use too much ram and make the os crash. Even though I decreased some parameter, it still use too much ram
Hello author, I'm a student who interested in your works when reading Multi-hop Attention Graph Neural Networks Can I get the information about the hardware that you train your model I cloned your repository and I on server in my lab which has Ubuntu 16.04, 32GB RAM
The code build well but it use too much ram and make the os crash. Even though I decreased some parameter, it still use too much ram
have you solved the problem? i met the same problem.
@lendie I solved the problem by using the smaller dataset. I read https://github.com/thiviyanT/torch-rgcn someday ago and I found this author use a subset of FB15k-237 called FB-toy. And I already test this code for it, and success.
@lendie I solved the problem by using the smaller dataset. I read https://github.com/thiviyanT/torch-rgcn someday ago and I found this author use a subset of FB15k-237 called FB-toy. And I already test this code for it, and success.
@nhutnamhcmus thanks a lot. I test this code for it too, but do you get the same result this paper shows?, I find that valid Hit@10 at step 28000 is too small(0.000824!)
@lendie I solved the problem by using the smaller dataset. I read https://github.com/thiviyanT/torch-rgcn someday ago and I found this author use a subset of FB15k-237 called FB-toy. And I already test this code for it, and success.
@nhutnamhcmus thanks a lot. I test this code for it too, but do you get the same result this paper shows?, I find that valid Hit@10 at step 28000 is too small(0.000824!)
@lendie I tested with hyperparameter like the paper said, and I setting for max step is 40,000. At step 20,000 I got the result Valid hits@10 0.500000 and valid MRR is 0.327864. So can you tell me the parameter you test
Do you think the time training is too long :) ?
@nhutnamhcmus I tested it on CPU，maybe it was one of the reasons I got so low(I did not alter the hyperparameter). I do think the time training is too long.