项目文件夹

文件

T

Chao Ma 86d60a1ff6 [DEMO] Add Pytorch demo for distributed sampler (#562 )

* update

* update

* update

* add sender

* update

* remove duplicate cpde

2019-05-24 11:53:28 +08:00

Stochastic Training for Graph Convolutional Networks

bash pip install torch requests

Test accuracy ~83% with --num-neighbors 2, ~84% by training on the full graph

DGLBACKEND=pytorch python3 gcn_ns_sc.py --dataset cora --self-loop --num-neighbors 2 --batch-size 1000000 --test-batch-size 1000000

Test accuracy ~69% with --num-neighbors 2, ~70% by training on the full graph

DGLBACKEND=pytorch python3 gcn_ns_sc.py --dataset citeseer --self-loop --num-neighbors 2 --batch-size 1000000 --test-batch-size 1000000

Test accuracy ~76% with --num-neighbors 3, ~77% by training on the full graph

DGLBACKEND=pytorch python3 gcn_ns_sc.py --dataset pubmed --self-loop --num-neighbors 3 --batch-size 1000000 --test-batch-size 1000000

Test accuracy ~84% with --num-neighbors 1, ~84% by training on the full graph

DGLBACKEND=pytorch python3 gcn_cv_sc.py --dataset cora --self-loop --num-neighbors 1 --batch-size 1000000 --test-batch-size 1000000

Test accuracy ~69% with --num-neighbors 1, ~70% by training on the full graph

DGLBACKEND=pytorch python3 gcn_cv_sc.py --dataset citeseer --self-loop --num-neighbors 1 --batch-size 1000000 --test-batch-size 1000000

Test accuracy ~77% with --num-neighbors 1, ~77% by training on the full graph

DGLBACKEND=pytorch python3 gcn_cv_sc.py --dataset pubmed --self-loop --num-neighbors 1 --batch-size 1000000 --test-batch-size 1000000