Code much slower than statement in paper

JD-AI-Research-Silicon-Valley / SACN

End-to-end Structure-Aware Convolutional Networks for Knowledge Base Completion

MIT License

113 stars 30 forks source link

Code much slower than statement in paper #3

Closed qiweizhen closed 5 years ago

qiweizhen commented 5 years ago

In the paper, there wrote that, 'On NVIDIA Tesla P40 GPU, for FB15k-237, computation time for SACN for each epoch is about 1 minute', but in default setting of your code and with a Tesla P100 GPU, it ran about 20 minutes 1 epoch. Did I set something wrong about your code? How many epoches would it take to converge?

qiweizhen commented 5 years ago

I analyzed the code with cProfile and it told me that, the GCN layer forward function took 99.6% time of the whole model forward time, and in GCN layer, tolist() function took more than half of the own time. And mul function and add_ function ranked the 2nd and 3rd most time cosuming functions.

chaoshangcs commented 5 years ago

Hi Qiwei, thank you so much for your messages and interests. We found a bug in the code which influences the running time. I am testing the new code now. Will upload it ASAP. Thanks!

qiweizhen commented 5 years ago

Thank you so much for your kindly reply

chaoshangcs commented 5 years ago

Dear Qiwei, I have updated a new version for testing. If you find any problem, please feel free and email me. Thanks!