Examples of Generative Model

[nashid]

🚀 Feature

Support for Generative Models

Motivation

DGL is intuitive to use and there are some great examples. However, DGL lacks generative models.

I am wondering whether there are any plans to include any of the following models:

• GraphRNN: https://github.com/JiaxuanYou/graph-generation

Also, DGL lacks examples of Graph2Seq based models. It would be awesome to consider any of the following Graph2Seq based generative models.

• Graph-to-Sequence Learning using Gated Graph Neural Networks, ACL’18 https://github.com/beckdaniel/acl2018_graph2seq Using MXNet

• Densely Connected Graph Convolutional Networks for Graph-to-Sequence Learning, ACL’19 https://github.com/Cartus/DCGCN Using MXNet 1.3.0

• Gated Graph Sequence Neural Networks, Y. Li, D. Tarlow, M. Brockschmidt, and R. Zemel.

• Heterogeneous Graph Transformer for Graph-to-Sequence Learning, ACL’18 https://github.com/QAQ-v/HetGT

Alternatives

I am considering the following GGNN implementation from OpenNMT.

OpenNMT GGNN https://opennmt.net/OpenNMT-py/examples/GGNN.html

Pitch

DGL is easy to use and mostly for predictive tasks. Supporting more generative models would clearly increase its adoption.

[nashid]

@sneakerkg OpenNMT has a GGNN implementation. https://opennmt.net/OpenNMT-py/examples/GGNN.html

I think having a similar model with DGL would be a great starting point!

[sneakerkg]

Hi @nashid , thanks very much for proposing this and adding the paper links. DGL team indeed has the plan and we are arranging force to add graph generative models, but the ETA may not be in this month. We will keep you posted if we have updates on this.

[jermainewang]

@nashid You are also very welcomed to contribute along this line. We are happy to discuss the details and provide help if needed.

[sneakerkg]

Hi @nashid, recently we've noticed a third-party library built upon DGL: https://github.com/graph4ai/graph4nlp . They have Graph2Seq implemented. Does that one include the models you want?

[nashid]

@sneakerkg thanks for pointing out this library. I am currently exploring Graph4NLP. Graph4NLP is a great library, but in my first attempt, I do not find it a generic library like DGL and quite hard to incorporate a custom dataset into their Graph2Seq model.

[nashid]

@sneakerkg wondering any further update?

[smith-co]

@sneakerkg so still no support for a generative model with DGL?

[jermainewang]

I've put GraphRNN to the roadmap issue #3849 so we can properly track them. For the Graph2Seq series of work, they all seem to years ago. Are there any more recent ones? Also cc @rudongyu @QipengGuo to see if you have any suggestions.

[smith-co]

@rudongyu @QipengGuo @jermainewang when do you plan to have a generative model like Graph2Seq?

[code-rex1]

@jermainewang @rudongyu @QipengGuo community has been waiting for this support for years, but there is no support yet for it from DGL 😢 😢 😢

Curious to know why is this case? Are GNNs not good enough for generative tasks?

[ramsey-coding]

Would be great to have an example of generative models like Graph2Seq like Graph2RNN, Graph2transformer, or GraphTransformer implementation. Wondering whether would it be added at some point?

[jermainewang]

Hi everyone, for graph transformer related work, we are closely watching and will likely add some supports such as examples or modules. Please stay tuned. We will post a dedicated roadmap for that soon!

[jermainewang]

Also, check out graph4nlp https://github.com/graph4ai/graph4nlp which contains many models like Graph2Seq.