Graph2Vec—— how the attributed nodes are transformed to subgraph and word?

[guyuleaves]

As the paper Graph2Vec presents in 4.3.1, each node will generate a subgraph using the WL kernel iteratively, where the node features are as utilized.

In my program, each node has a feature list of 3 length, like node1:[0.5,0.2,0.3], and i set the args attributed TRUE of graph2vec. In contrary, i find that if attributed is set FALSE, the node feature is given its degree.

However, in the code of WeisfeilerLehmanHashing, the nodes are transformed to a hash MD5 code directly, where i don't understand that how the node and its adjacent nodes are combinded to subgraph and then encoded to a word? Especially how the node feature list are processed in subgraph and word generation.