Neighbor Sampling & Skip Connection

Models & Apps
#1

When “python gcn_ns_sc.py --dataset cora --self-loop --num-neighbors 2 --batch-size 1000000 --test-batch-size 1000000 --gpu 0” is being executed, I’m experiencing RuntimeErrorL Expected object of backend CUDA but got backend CPU for argument #4 ‘mat1. Why？

#2

@zhengda1936 Could you please take a look?

#3

IMG_20190614_154636
IMG_20190614_1546362660×1427 2.38 MB

#4

I think this is a bug in our 0.2 release. We’ve released v0.3 yesterday, could you try our new version and see whether this error still exists? Thanks

#5

0.3 installation guide: https://docs.dgl.ai/install/index.html
Please pay attention to cuda installation guide if you want to use gpu.

#6

I want to install DGL 0.3.
According to 0.3 installation guide, I know how to install DGL for CUDA 9.0 and CUDA 10.0, but for CUDA 9.2 build, how to run?

#7

@cornelia Sorry we don’t have cu92 build release. Will try to upload it soon.

#8

Hi @cornelia, we just pushed the Windows and Linux CUDA 9.2 build to anaconda. Please give it a try.

conda install dgl-cuda9.2 -c dglteam