Trainning problems for ppi_bp dataset

[youngfish42]

I want to repeat experiment results, and when try to use ppi_bp dataset to train the model,

python train_config.py -config_path config_files/ppi_bp/ppi_bp_config.json

I get the following message:

Running 50 Trials of optuna
Logging to  /data/SubGNN-datasets/tensorboard/S_ppi_bp_optuna
[I 2021-07-27 16:54:48,142] Using an existing study with name '/data/SubGNN-datasets/tensorboard/S_ppi_bp_optuna' instead of creating a new one.
[9339]
--- Finished reading in data ---
Tensorboard logging at  /data/SubGNN-datasets/tensorboard/S_ppi_bp_optuna/version_8722918
--- Started Preparing Data ---
--- Initializing CC Embeddings ---
--- Initializing CC Border Sets ---
--- Getting Similarities ---
--- Loading Train Structure Similarities from File ---
--- Loading Val Structure Similarities from File ---
Done computing internal structure similarities
computing border structure sims
---Computing Structure Patch Similarities---
/data/usr/yangyuwen/SubGNN/SubGNN/gamma.py:53: RuntimeWarning: divide by zero encountered in double_scalars
  return ((max(a,b) + 1)/(min(a,b) + 1)) - 1
/data/usr/yangyuwen/SubGNN/SubGNN/gamma.py:53: RuntimeWarning: divide by zero encountered in double_scalars
  return ((max(a,b) + 1)/(min(a,b) + 1)) - 1
/data/usr/yangyuwen/SubGNN/SubGNN/gamma.py:53: RuntimeWarning: divide by zero encountered in double_scalars
  return ((max(a,b) + 1)/(min(a,b) + 1)) - 1
/data/usr/yangyuwen/SubGNN/SubGNN/gamma.py:53: RuntimeWarning: divide by zero encountered in double_scalars
  return ((max(a,b) + 1)/(min(a,b) + 1)) - 1

And the program will be stuck here, maybe more than 24 hours. (I stopped it after that)

[Isabellajhon]

@youngfish42 Have you figured it out?

[youngfish42]

@youngfish42 Have you figured it out?

Sorry, I have never successfully reproduced the experiment using the PPI_BP data set. For this reason, we removed the experiment on this data set in the subsequent paper: Position-Aware Subgraph Neural Networks with Data-Efficient Learning.

But what I want to add is that other teams’ follow-up papers: GLASS: GNN with Labeling Tricks for Subgraph Representation Learning mentioned experiments on this set of data sets. You may be able to contact them for solutions.