Over-squash

---

Description

Over-squash is a tool for solving the problem of information bottlenecks in graph neural networks (GNNs) or implementing effective mechanisms to mitigate over-squashing issues. This project uses Python and deep learning frameworks, including PyTorch and PyTorch Geometric. It aims to enhance the ability of GNNs to handle long-range dependencies without suffering from information loss or compression, thereby improving their performance in tasks requiring deep relational information.

---

Installation For Over-squashing and transductive learning

To set up the project environment and install all necessary dependencies, follow these steps:

1. Clone the repository:

   git clone https://github.com/yonatansverdlov/Over-squashing.git

2. Navigate into the project directory:

   cd Over-squashing

3. Create a new Conda environment and activate it:

   conda create --name oversquash -c conda-forge python=3.11
   conda activate oversquash

4. Install the necessary dependencies from the requirements.txt file:

   pip install -r requirements.txt

---

Usage

We present three types of experiments: Over-squashing experiments, Transductive learning, MolHIV and LRGB.

Over-squashing experiments

First run

cd bottleneck/script

Choose data_type, one of the four options: Ring, Tree, CrossRing, CliqueRing. Then, for Tree, choose a radius between 2 and 8, and for others, between 2 and 15.

If all radios are needed, please run

python train.py --dataset_name data_type --all True

Otherwise, run

python train.py --dataset_name data_type --radius radius.

---

Transductive Learning

First run

cd bottleneck/script

Select a data_type from the following nine options: Cora, Cite, Pubm, Cham, Squi, Actor, Corn, Texas, Wisc. Next, choose the number of different seeds (between 1 and 10) indicated by repeat, and run:

python train.py --dataset_name data_type --repeat repeat

LRGB & MolHIV

1. Create a new Conda environment and activate it:

   cd Over-squashing
   conda create --name lrgb -c conda-forge python=3.10
   conda activate lrgb

2. Install the necessary dependencies from the lrgb_requirements.txt file:

   pip install -r lrgb_requirements.txt

License

This project is licensed under the MIT License.

---

Contact

For any questions or feedback, reach out to me at [email protected].