kgcnn 1.1.0

Changes, updates and bug fixes

• New training scripts which are model independent and serializable.
• Trainig scripts with command line parameters to load hyper-parameter dictionary from json-file.
• Modified model input parameters for all models.
• Default hyper-parameters per dataset.
• Updated documentation in several modules.
• Reorganized layers in kgcnn.layers.conv and kgcnn.layers.pool.
• Minor bug fixes.

kgcnn 1.0.2

Changes, updates and bug fixes

• New models: GIN, PAiNN, GATv2 in kgcnn.literature
• Refactor model args and datasets.
• Remove disjoint list format, simplify layer base class.

kgcnn 1.0.1

Changes, updates and bug fixes

• Fix bug in DimeNet++ and training script from QM9.
• Rework dataset loading via dataset class .
• Add AttentiveFP with training script.
• Enable serialization for most models.

kgcnn 1.0.0

Realse version 1.0

Graph neural networks in TensorFlow-Keras with RaggedTensor representation kgcnn. The package in kgcnn contains several keras layer classes to build up graph convolution models. Some models from literature are given as an example.

kgcnn 0.1.3

Changes, updates and bug fixes

• Merged layer and added DimNet.
• Further layer rework is planned by inheriting from GraphBaseLayer

kgcnn 0.1.1

Changes, updates and bug fixes

• Refactor model and layer names.
• Added GNNExplain.

kgcnn 0.1.0

Initial release

Draft for graph neural networks with tf.keras.