fedml-ai / fedgraphnn

FedGraphNN: A Federated Learning Platform for Graph Neural Networks with MLOps Support. The previous research version is accepted to ICLR'2021 - DPML and MLSys'21 - GNNSys workshops.

FedGraphNN has been migrated to https://github.com/FedML-AI/FedML/tree/master/python/app/fedgraphnn.

fedgraphnn's Issues

The distributed experiment was stuck after creating model done

I ran run_fedavg_distributed_pytorch but the experiment was stuck after creating the model done. What's wrong?

2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 0, local sample number = 191
2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 1, local sample number = 190
2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 2, local sample number = 190
2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 3, local sample number = 190
2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 4, local sample number = 190
2022-04-10,23:37:38.903 - {data_loader.py (453)} - load_partition_data(): Client idx = 5, local sample number = 190
2022-04-10,23:37:38.904 - {main_fedavg.py (139)} - create_model(): create_model. model_name = graphsage, output_dim = None
2022-04-10,23:37:38.929 - {main_fedavg.py (180)} - create_model(): done

Link does not work

This link for FedGraphNN does not work:
https://github.com/FedML-AI/FedML/tree/master/python/app/fedgraphnn

terminate called after throwing an instance of 'std::bad_alloc'

Hi, I encountered the 'std::bad_alloc' problem when running ego-networks problems, any suggestions for debugging?
terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc

Thanks.

AttributeError: 'Namespace' object has no attribute 'backend'

I'm trying to follow the classification experiments:
From this link: https://github.com/FedML-AI/FedGraphNN/tree/main/experiments/distributed/moleculenet
This command results in the error:
sh run_fedavg_distributed_pytorch.sh 4 4 1 4 graphsage hetero 0.2 20 1 1 0.0015 64 32 0.3 64 64 sider "./../../../data/sider/" 0

About the missing code

Hellow!
I am a beginner in knowledge graph embedding. It's a very up-to-date work which help a lot for me to understand the task of knowledge graph reasoning and federated learning. But when I try to run the relation prediction code,I find the 'fed_subgraph_rel_trainer' is missing. And because of the access problem,I can't download the processed datasets in the subgraph-level like FB15k-237 and WN18RR from the web you provided.Would you mind share these datasets and the missing file ''fed_subgraph_rel_trainer' ?

Thank you very much for the early reply.

illegal hardware instruction python

I'm using m1 pro chip. It seems that when I try to run centralized experiment it will report illegal hardware instruction. Is there any solution to it? thanks

post_complete_message_to_sweep_process in main_fedavg.py is useless.

In experiments/distributed/moleculenet/main_fedavg.py, several lines of codes at the end of the file are shown as follows:

if process_id == 0:
        post_complete_message_to_sweep_process(args)

Actually, these two lines of codes cannot be run, since before getting into them the application will call MPI_Abort to terminate the processes.

[Missing file] No module named 'training.subgraph_level.fed_subgraph_rel_trainer'

Hi!

When trying to run the code of "subgraph relation prediction", I received this feedback:

Traceback (most recent call last):
File "fed_subgraph_rel_pred.py", line 17, in
from training.subgraph_level.fed_subgraph_rel_trainer import FedSubgraphRelTrainer
ModuleNotFoundError: No module named 'training.subgraph_level.fed_subgraph_rel_trainer'

I find there is no file named "fed_subgraph_rel_trainer" under the folder "training/subgraph_level" . And the README of this part is the same as "Ego Networks".

Hope you can complete the file. Thanks!

Cannot load "ciao" and "epinions" datasets

When loading the "ciao" and "epinions" datasets with function "load_partition_data", there is an error as follows:

RuntimeError: The 'data' object was created by an older version of PyG. If this error occurred while loading an already existing dataset, remove the 'processed/' directory in the dataset's root folder and try again.

Is there any solution to this error?

A typo？

For Installation

conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch -n **fedmolecule**
The parameter ‘fedmolecule’ maybe 'fedgraphnn'

invalid value for int argument: NODE_DIM

I am trying to run the Distributed Molecule classification experiment:
sh run_fedavg_distributed_reg.sh 6 1 1 1 graphsage homo 150 1 1 0.0015 256 256 0.3 256 256 freesolv "./../../../data/freesolv/" 0

Environment settings for ego-networks

Are the environment settings for ego-networks and moleculenet the same? I can run distributed experiments for moleculenet but fail to run that of ego-networks under the same environment.

About data

Hi There
Would you mind sharing how to get .pkl files and .npy files downloaded in data directory shell script?
Thanks a lot!

error: argument --node_embedding_dim: invalid int value: '0.3'

Just try to run the experiment based on the command as follow:
sh run_fedavg_distributed_pytorch.sh 6 1 1 1 graphsage homo 150 1 1 0.0015 256 256 0.3 256 256 sider "./../../../data/sider/" 0