godm

KO | EN

gitlite — search

#javascript #typescript #machine-learning #rust #docker #macos #nodejs #windows #kubernetes #awesome-list #openai #mcp #ios #open-source

(★ 15)

Data Augmentation for Supervised Graph Outlier Detection with Latent Diffusion Models

godm Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# GODM

[![PyPI version](https://badge.fury.io/py/godm.svg)](https://badge.fury.io/py/godm)

GODM is a data augmentation package for supervised graph outlier detection. It generates synthetic graph outliers with latent diffusion models. This is the official implementation of [Data Augmentation for Supervised Graph Outlier Detection with Latent Diffusion Models](https://arxiv.org/abs/2312.17679).

## Installation

It is recommended to use **pip** for installation:

```pip install godm```

Alternatively, you can build from source by cloning this repository:

```
git clone https://github.com/kayzliu/godm.git
cd pygod
pip install .
```

If you have trouble installing `metis`, take a look at [this tutorial](https://medium.com/@beppe2hd/a-simple-guide-to-set-up-pyg-with-metis-support-on-mac-and-linux-378e833d4de3).

## Usage

```python
from pygod.utils import load_data
data = load_data('weibo') # load data

from godm import GODM     # import GODM
godm = GODM(lr=0.004)     # init. GODM
aug_data = godm(data)     # augment data

detector(aug_data)        # train on data
```

The input data should be [`torch_geometric.Data`](https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.data.Data.html#torch_geometric.data.Data) object with the following keys:

- `x`: node features,
- `edge_index`: edge index, 
- `edge_time`: edge times (optional, name can be changed by `time_attr`),
- `edge_type`: edge types (optional, name can be changed by `type_attr`), 
- `y`: node labels, 
- `train_mask`: training node mask, 
- `val_mask`: validation node mask, 
- `test_mask`: testing node mask.

So far, no additional keys is allowed. We may support more keys by padding in the future.

## Parameters

- ```hid_dim``` (type: `int`, default: `None`): hidden dimension for VAE, i.e., latent embedding dimension. `None` means the largest power of 2 that is less than or equal to the feature dimension divided by two.
- ```diff_dim``` (type: `int`, default: `None`): hidden dimension for denoiser. `None` means as twice as `hid_dim`.
- ```vae_epochs``` (type: `int`, default: `100`): number of epochs for training VAE.
- ```diff_epochs``` (type: `int`, default: `100`): number of epochs for training diffusion model.
- ```patience``` (type: `int`, default: `50`): patience for early stopping.
- ```lr``` (type: `float`, default: `0.001`): learning rate.
- ```wd``` (type: `float`, default: `0.`): weight decay.
- ```batch_size``` (type: `int`, default: `2048`): batch size.
- ```threshold``` (type: `float`, default: `0.75`): threshold for edge generation.
- ```wx``` (type: `float`, default: `1.`): weight for node feature reconstruction loss.
- ```we``` (type: `float`, default: `0.5`): weight for edge reconstruction loss.
- ```beta``` (type: `float`, default: `0.001`): weight for KL divergence loss.
- ```wt``` (type: `float`, default: `1.`): weight for time prediction loss.
- ```time_attr``` (type: `str`, default: `edge_time`): attribute name for edge time.
- ```type_attr``` (type: `str`, default: `edge_type`): attribute name for edge type.
- ```wp``` (type: `float`, default: `0.3`): weight for node prediction loss.
- ```gen_nodes``` (type: `int`, default: `None`): number of nodes to generate. `None` means the same as the number of outliers in the original graph.
- ```sample_steps``` (type: `int`, default: `50`): number of steps for diffusion model sampling.
- ```device``` (type: `int`, default: `0`): GPU index, set to -1 for CPU.
- ```verbose``` (type: `bool`, default: `False`): verbose mode, enable for logging.

## Cite Us:

Our [paper](https://arxiv.org/abs/2312.17679) is publicly available. If you use GODM in a scientific publication, we would appreciate your citations:

@InProceedings{pmlr-v269-liu25b,
      title={Data Augmentation for Supervised Graph Outlier Detection via Latent Diffusion Models},
      author={Liu, Kay and Zhang, Hengrui and Hu, Ziqing and Wang, Fangxin and Yu, Philip S.},
      booktitle={Proceedings of the Third Learning on Graphs Conference},
      pages={43:1--43:20},
      year={2025},
      editor={Wolf, Guy and Krishnaswamy, Smita},
      volume={269},
      series={Proceedings of Machine Learning Research},
      month={26--29 Nov},
      publisher={PMLR},
      pdf={https://raw.githubusercontent.com/mlresearch/v269/main/assets/liu25b/liu25b.pdf},
      url={https://proceedings.mlr.press/v269/liu25b.html}
    }

or:

Liu, Kay, Hengrui Zhang, Ziqing Hu, Fangxin Wang, Philip S. Yu. "Data Augmentation for Supervised Graph Outlier Detection via Latent Diffusion Models." Proceedings of the Third Learning on Graphs Conference, PMLR 269:43:1-43:20, 2025.
    
##

godm

godm Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

godm

godm Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README