Skip to content

Latest commit

 

History

History
89 lines (64 loc) · 3.11 KB

README.md

File metadata and controls

89 lines (64 loc) · 3.11 KB
Raw

grafog

Graph Data Augmentation Library for PyTorch Geometric.

What is it?

Data augmentations are heavily used in Computer Vision and Natural Language Processing to address data imbalance, data scarcity, and prevent models from overfitting. They have also proven to yield good results in both supervised and self-supervised (contrastive) settings.

grafog (portmanteau of "graph" and "augmentation") provides a set of methods to perform data augmentation on graph-structured data, especially meant for self-supervised node classification. It is built on top of torch_geometric and is easily integrable with its Data API.

Yannic Kilcher talks about it here: https://youtu.be/smUHQndcmOY?t=961

Installation

You can install the library via pip:

$ pip install grafog

You can also install the library from source:

$ git clone https://github.com/rish-16/grafog
$ cd grafog
$ pip install -e .

Dependencies

torch==1.10.2
torch_geometric==2.0.3

Usage

The library comes with the following data augmentations:

Augmentation Remarks When to use
NodeDrop(p=0.05) Randomly drops nodes with the given p before, during training
EdgeDrop(p=0.05) Randomly drops edges with the given p before, during training
Normalize() Normalizes the node or edge features before training
NodeMixUp(lamb, classes) MixUp on node features with given lambda during training
NodeFeatureMasking(p=0.15) Randomly masks node features with the given p during training
EdgeFeatureMasking(p=0.15) Randomly masks edge features with the given p during training

There are many more features to be added over time, so stay tuned!

from torch_geometric.datasets import CoraFull
import grafog.transforms as T

node_aug = T.Compose([
    T.NodeDrop(p=0.45),
    T.NodeMixUp(lamb=0.5, classes=7),
    ...
])

edge_aug = T.Compose([
    T.EdgeDrop(0=0.15),
    T.EdgeFeatureMasking()
])

data = CoraFull()
model = ...

for epoch in range(10): # begin training loop
    new_data = node_aug(data) # apply the node augmentation(s)
    new_data = edge_aug(new_data) # apply the edge augmentation(s)
    
    x, y = new_data.x, new_data.y
    ...

Remarks

This library was built as a project for a class (UIT2201) at NUS. I planned and built it over the span of 10 weeks. I thank Prof. Mikhail Filippov for his guidance, feedback, and support!

If you spot any issues, feel free to raise a PR or Issue. All meaningful contributions welcome!

License

MIT