search

microsoft / constrained-graph-variational-autoencoder

Sample code for Constrained Graph Variational Autoencoders
MIT License
231 stars 57 forks source link

Tensorflow2 implementation #6

Open YeolYao opened 2 years ago

YeolYao commented 2 years ago

Hello, thank you for this great paper and great implementation! As Tensorflow has upgraded to version 2, I was wondering if you by any chance changed this model to Tensorflow2 as well?

leuchine commented 2 years ago

Hi,

Thanks for your interest. However, this repo is not actively maintained anymore. It is better to use the older version of tensorflow. Otherwise, some functions may be broken.

YeolYao commented 2 years ago

Hello, I have some questions to ask. I noticed in your paper you said that the model only considers undirected graphs, is it possible to extend it to directed graphs? The current model is used for molecular graph generation, so the node features and edge features are only onehot, is it possible to extend other numerical features? Looking forward to your reply, Thanks a lot!

leuchine commented 2 years ago

Hi,

Extending to directed graphs is easy. Need to maintain the direction of edges during expansion.

Using numerical features and edge features is not a problem. But I think it requires a network to output continuous values instead of performing a classification in the code