parthapratimbanik / pygod Goto Github PK

Using the package to detect edge level and sub-graph level anomalies.

To Do:

• Adding an ipynb style example for edge level or (sub-)graph level outlier detection

Issue of pygod-team#83

Describe the bug

The results on inj_cora (AUC: 0.7566±0.0332 (0.7751)) and inj_amazon (AUC: 0.7147±0.0006 (0.7152)) are significantly different from what you show in Table 3 from the BOND paper (https://arxiv.org/pdf/2206.10071.pdf), which are 82.7±5.6 (84.3) on inj_cora and 81.3±1.0 (82.2) for inj_amazon.

Inconsistency Result of BOND paper, Table 3 (on inj_cora and inj_amazon datasets)

To Reproduce
Steps to reproduce the behavior:
May be error found in pygod v1.0.0

Expected behavior
AUC result must be alike Table 3 of the BOND paper (https://arxiv.org/pdf/2206.10071.pdf)
By downgrading to the original version used by the benchmark v0.3.1 via pip install pygod=0.3.1, the expected results can be regenerated.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: [e.g. iOS]
• Browser [e.g. chrome, safari]
• Version [e.g. 22]

To Do
According to @kayzliu, the following things need to do.

• Have to update model initialization in utils.
• Remove heuristic selection of weight.
• Update the parameter name for detectors (DOMINANT, GAAN, and CONAD) from alpha to weight.

Have to implement the following methods:

• def __init__(...): it initializes the model
• def forward(...): it executes the model forward operation
• def configure_optimizers(...): it configures optimizer
• def training_step(...): it executes on every training steps + on each epoch
• Must be equal or close to v1.0.0 DOMINANT + inj_cora result

Title:
Wrapping with Pytorch Lightning

Description:
We would like to start improving the model scalability via Pytorch Lightning (https://www.pytorchlightning.ai/)

To Do:
by using Pytorch Lightning:

• #4
• #5
• Do the above steps for other models
  • will be updated later

Have to implement the following things by using torch.utils.data.Dataset:

• def __init__(...): Setup data directory
• def __len__(...): Return the length of sample
• def __getitem__(...): Return the sample (x, y)
  
  Have to Implement the following things by using LightningDataModule, for the inj_cora dataset:
• def __init__(...): setup data directory
• def preapre_data(...): download data in local
• def setup(...): load data from local
• def train_dataloaders(...): setup training data for training
• Must be equal or close to load_data() result for inj_cora dataset
  
  Dataset preparations depend on the following functions and variables:

1. to_dense_adj(...)
2. num_neigh + num_layers
3. NeighborLoader(...)
   
   The above factors are dependent on models. So, have to do the following things:

• Point out models with to_dense_adj(...) and/or num_neigh + num_layers and NeighborLoader(...)

Have to do the following things, based on the above implementations:

• Comparing the output of the following datasets with the load_data() output:
  • inj_amazon
  • inj_flickr
  • weibo
  • reddit
  • disney
  • books
  • enron