21S003018 / FL-AGCNS

This branch is for cora citeseer pubmed corafull Physics, if running SBM goto SBM

Environment

soft：torch_geometric

hard：many GPUs

Run Federated Learning model

Let's take cora dataset and our fl model as example.

nohup python -u server.py --client 3 --model fl-agcns --dataset cora 2>&1 >> log/fl-agcns_sbm_server.log &
nohup python -u client.py --client 3 --model fl-agcns 2>&1 >> log/fl-agcns_sbm_client.log &

Please open two terminals, input the first command line into one terminal and input the second one into another terminal.
And the concete commands are as follows

server end

usage: server.py [-h] [--mode MODE]
                 [--code SUPERMASK SUPERMASK SUPERMASK SUPERMASK SUPERMASK SUPERMASK SUPERMASK SUPERMASK]
                 [--model {fl-darts,fl-agcns,fl-random,fl-graphnas,fl-fednas}]
                 [--dataset {citeseer,Physics,pubmed,corafull,cora}]
                 [--client CLIENT]

client end

usage: client.py [-h] [--mode MODE]
                 [--model {fl-darts,fl-agcns,fl-random,fl-graphnas,fl-fednas}]
                 [--client CLIENT]

Example

We have searched some good architectures and you could evaluate it using the following command.

# cora
python server.py --mode test --code 1 6 0 0 6 54 55 5 --dataset cora
python client.py --mode test
# citeseer
python server.py --mode test --code 1 4 18 30 42 0 0 5 --dataset citeseer
python client.py --mode test
# pubmed
python server.py --mode test --client 12 --code 1 0 0 6 42 43 0 5 --dataset pubmed
python client.py --mode test --client 12
# corafull(extension to cora, you could use cora's code and the presented in paper is 1 6 0 6 42 54 55 5)
python server.py --mode test --client 16 --code 1 6 0 0 6 54 55 5 --dataset pubmed
python client.py --mode test --client 16
# Physics
python server.py --mode test --client 18 --code 1 5 0 7 0 42 0 4 --dataset Physics
python client.py --mode test --client 18

Please open two terminals, input the first command line into one terminal and input the second one into another terminal.

Note

1. Maybe you should run python make_data.py first to prepare the dataset.
2. If connection refused，please run reset.py to unify the ip_port.
3. run server.py before client.py