Spektral

Seamlessly integrates with the Keras API and TensorFlow 2, allowing developers familiar with Keras to easily build and train graph neural networks using familiar workflows, as highlighted in the README's emphasis on compatibility.

Implements a wide range of popular GNN layers, including GCN, GAT, GraphSAGE, and GIN, providing flexibility for various graph tasks without needing to code from scratch, as listed in the convolutional and pooling layers sections.

The new Dataset API, Graph containers, and Loader class simplify data representation, transformation, and batching, making it easier to handle graph-structured data in deep learning projects, as described in the 'New in Spektral 1.0' section.

Offers GeneralConv and GeneralGNN classes with state-of-the-art defaults, enabling versatile model creation that works well across different tasks without extensive tuning, as mentioned in the key features.