Benchmarking-Loss-Functions

About the project

We tested a number of local structure preserving loss functions and their influence on node classification task.

Installation

1. Run pip install -r requirements.txt
2. Install Torch 1.11.0
3. Install Torch Geometric in current experiments PyG==2.1.0 is used.

How to use

examples are in NodeClassificaton.jpynb

Available loss functions

In some loss functions we added negative sampling optimization (VERSE, LINE, DeepWalk, Node2Vec)

• Laplacian Eigenmaps
• Graph Factorization
• HOPE
• VERSE
• LINE
• APP
• DeepWalk
• Node2Vec
• Force2Vec

Available convolutions

• GCN
• GAT
• SAGE