refactoring for keras 3.0

docs/source/index.rst

@@ -6,7 +6,11 @@
 General Information
 ===================================
 
-A set of layers for graph convolutions in TensorFlow Keras that use RaggedTensors.
+The package in `kgcnn <https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn>`__ contains several layer classes to build up graph convolution models in
+Keras with Tensorflow, PyTorch or Jax as backend. Some models are given as an example in literature.
+Focus of kgcnn is (batched) graph learning for molecules **kgcnn.molecule** and materials **kgcnn.crystal**.
+Below you can find explanations and information on how to use `kgcnn <https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn>`__ .
+See `Reference` under `Package Content` for code documentation.
 
 .. toctree::
    :maxdepth: 3

docs/source/installation.rst

@@ -4,20 +4,14 @@
 Installation
 ============
 
-Clone repository https://github.com/aimat-lab/gcnn_keras and install with editable mode::
-
-   pip install -e ./gcnn_keras
-
+Clone repository https://github.com/aimat-lab/gcnn_keras and install with editable mode
 or latest release via `Python Package Index <https://pypi.org/>`_ ::
 
    pip install kgcnn
 
-Standard python package requirements are placed in the ``setup.py`` and are installed automatically (`kgcnn <https://github.com/aimat-lab/gcnn_keras>`_ >=2.2).
-Packages which must be installed manually for full functionality are listed below.
-For example to convert molecular file formats `OpenBabel <http://openbabel.org/wiki/Main_Page>`_ can complement `RDkit <https://www.rdkit.org/docs/api-docs.html>`_ which is installed via ``pip``.
-
-* `OpenBabel <http://openbabel.org/wiki/Main_Page>`_ >=3.0.1
 
-To have proper GPU support, make sure that the installed ``tensorflow`` version matches your system requirements.
+Standard python package requirements are placed in the ``setup.py`` and are installed automatically.
+However, you must make sure to install the GPU/TPU acceleration for the backend of your choice.
+For example, to have proper GPU support, make sure that the installed ``tensorflow`` version matches your system requirements.
 Moreover, installed `GPU drivers <https://www.nvidia.com/download/index.aspx?lang=en-us>`_ and `CUDA <https://developer.nvidia.com/cuda-toolkit-archive>`_  and `cuDNN <https://developer.nvidia.com/cudnn>`_ versions must match.
 A list of verified version combinations can be found here: https://www.tensorflow.org/install/source#gpu .

docs/source/intro.rst

@@ -5,7 +5,12 @@ Introduction
 ============
 
 
-The package in `kgcnn <https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn>`_ contains several layer classes to build up graph convolution models.
+The package in `kgcnn <https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn>`__ contains several layer classes to build up graph convolution models.
 Some models are given as an example.
-A documentation is generated in `docs <https://github.com/aimat-lab/gcnn_keras/tree/master/docs>`_.
-Any comments, suggestions or help are very welcome!
+
+The version of **kgcnn>=4.0.0** are developed for **keras>=3.0** and should enable graph neural networks with multi-backend support.
+The currently supported backends are `PyTorch <https://pytorch.org/>`__ , `TensorFlow <https://www.tensorflow.org/install>`__ and `Jax <https://jax.readthedocs.io/en/latest/installation.html>`__ .
+
+All models and layers of kgcnn
+
+

kgcnn/layers/mlp.py

@@ -328,10 +328,10 @@ def call(self, inputs, **kwargs):
         r"""Forward pass.
 
         Args:
-            inputs (tf.Tensor): Input tensor with last dimension not `None` .
+            inputs (Tensor): Input tensor with last dimension not `None` .
 
         Returns:
-            tf.Tensor: MLP forward pass.
+            Tensor: MLP forward pass.
         """
         x, batch = (inputs[0], inputs[1:]) if isinstance(inputs, list) else (inputs, [])
 

kgcnn/layers/norm.py

@@ -286,7 +286,7 @@ class GraphInstanceNormalization(GraphNormalization):
     """
 
     def __init__(self, **kwargs):
-        r"""Initialize layer :obj:`GraphBatchNormalization`.
+        r"""Initialize layer :obj:`GraphBatchNormalization` .
 
         Args:
             epsilon: Small float added to variance to avoid dividing by zero. Defaults to 1e-3.