Featurizer for Graph Convolution NN?

Hello all, new here.

It is my understanding that the GraphConv featurizer transforms the molecular structure into a representation that is suitable for input into the GCNN model. After applying the transformation, the input data becomes a ConvMol python object.

My question is: What does the GraphConv featurizer actually do? What is the structure of the data once it is featurized and how does it represent the molecule?

I have looked at the tutorials and read the Duvenaud paper but can’t seem to find an answer. If someone can explain or direct me to a helpful resource, it would be greatly appreciated. Thanks!