DA-GDA: Differentiable Automatic Graph Data Augmentation for Semi-Supervised Node Classification

This repository is the official implementation of DA-GDA.

Requirements

To install requirements:

pip install -r requirements.txt

Training

To train and eval the model in the paper, run commands:

• python train_search.py
• python train_search_twitch.py

Configurations

We provide model configurations for each dataset under the configs directory.

Results

Our model achieves the following performance on:

1. Node classification accuracy / micro $F_1$ scorecomparison:

   

2. Efficiency evaluation:

   

3. Robustness evaluation:

   

4. Hyperparameter sensitivity evaluation:

   

5. Ablation study: