anasrz/cora-node-gcn

This is a NodeClassifier Graph Neural Network model on Cora built with K3-Node and Keras 3.

It runs natively and seamlessly across PyTorch, JAX, and TensorFlow backends.

Model Details

• Task: NodeClassifier
• Backbone Architecture: gcn
• Library: k3-node (Keras 3)
• Input Channels: 1433
• Hidden Channels: 32
• Output / Classes: 7
• Number of Layers: 2
• Dropout: 0.5

Evaluation Metrics

Metric	Value
accuracy	0.7860

Usage

Install k3-node with your preferred backend (PyTorch, JAX, or TensorFlow):

pip install k3-node huggingface_hub

Loading & Inference

import os
os.environ["KERAS_BACKEND"] = "torch"  # or "jax", "tensorflow"

from k3_node.tasks import NodeClassifier

# Load the pretrained model directly from Hugging Face Hub
model = NodeClassifier.from_pretrained("anasrz/cora-node-gcn")

# Run predictions on your graph data
predictions = model.predict(data)

Framework & Citation

This model was trained with K3-Node, the multi-backend Graph Neural Network framework built on Keras 3.

@software{k3_node,
  author = {Muhammad Anas Raza},
  title = {K3-Node: Multi-Backend Graph Neural Networks on Keras 3},
  year = {2026},
  url = {https://github.com/anas-rz/k3-node}
}