Closed BearBiscuit05 closed 11 months ago
Hi, thank you for your interest in our work!
You need to provide the cache allocation plan and other params to the run_ducati.py script
CUDA_VISIBLE_DEVICES=0 python run_ducati.py --dataset ogbn-papers100M --fanouts 10,25 --fake-dim 128 --adj-budget 0.142 --nfeat-budget 0.858
Hi, with your help I successfully ran the results, thank you very much. After studying the source code, I would like to ask, did you obtain the results by replacing the original neighbor sampling function in dgl with the CSRRowWiseSamplingUniformWithCache
function? Are there any other optimizations?
Hi, there are two major differences between DUCATI and previous cache works. (1) propose the adj cache design and thus support sampling with cache as you mentioned (2) develop a dual-cache algorithm to solve the cache contention problem.
The first part is implemented in both DUCATI and customized DGL, the second part is implemented only in DUCATI’s code. If you simply substitute the neighbor sampling function, the code cannot work . You can find the details about the contributions in our paper.
Thank you very much for your response, I have understood.
Hello, following the instructions in the README, I successfully configured the environment. However, when I tried to run PA according to the commands, it seems that some issues have occurred. I would like to know where the problem is or if there might be an issue with the parameters I entered?
Then