gnn

Let's implement ML models for semi-supervised learning.

Setting up environment

We are using uv to manage project dependencies.

# Install uv
$ curl -LsSf https://astral.sh/uv/install.sh | sh
# check uv version
$ uv --version
uv 0.8.3 (7e78f54e7 2025-07-24)

Create virtual environment using python 3.11.x.

$ uv venv --python 3.11
Using CPython 3.11.10
Creating virtual environment at: .venv
Activate with: source .venv/bin/activate

Install dependencies from pyproject.toml.

$ uv sync

If you added any packages to pyproject.toml, please run following command to sync dependencies.

$ uv lock

Setting up git hook

Set up automatic linting using the following commands:

# This command will ensure linting runs automatically every time you commit code.
pre-commit install

How to run experiment

Here is a sample code for running Two layers GCN on cora dataset.

$ export PYTHONPATH=.
$ uv run python3 gnn/train.py \
  --data_name cora \
  --model_name gcn \
  --epochs 200

After finishing training, all the results, such as loss, accuracy, best torch weight will be saved in result/untest/{model_name}/{dt}.

Experiment results

Model	Dataset	Test loss	Test Accuracy	Test Macro F1
GCN	cora	0.7404	0.7819	0.762
Graphsage (mean aggregator)	cora	0.8424	0.7302	0.705
GAT (head=1)	cora	0.9679	0.6873	0.6068
FastGCN (importance sampling)	cora	1.352	0.6359	0.6366

How to run pytest

Run following command.

$ uv run pytest