search

graph4ai / graph4nlp

Graph4nlp is the library for the easy use of Graph Neural Networks for NLP. Welcome to visit our DLG4NLP website (https://dlg4nlp.github.io/index.html) for various learning resources!
Apache License 2.0
1.67k stars 201 forks source link

Parameters for hyperparameter tuning #553

Closed code-rex1 closed 2 years ago

code-rex1 commented 2 years ago

❓ Questions and Help

Here you have presented a Graph2Seq and Graph2Tree model. I would like to know what are the search space I should use for these Graph2Seq and Graph2Tree model.

Would be really interested to know what should be the reasonable values for different graph variants as well.

I had look into your paper Graph4NLP survey to find out the reasonable value for tuning. Unfortunately, the survey does not shed light into this.

Would really appreciate your feedback about reasonable values for the hyper parameters?

schenglee commented 2 years ago

This is a nice question, we have put this issue into our TODO list, and we will soon build a document that could help hyperparameter tuning, stay tuned, please.

code-rex1 commented 2 years ago

@schenglee wondering do you have any update on this issue? 🙏

schenglee commented 2 years ago

Please refer here: https://docs.google.com/spreadsheets/d/e/2PACX-1vQaE3BTKYt4NX0z5oJrzVESdE7Kx3dnmTCG7zTdtTqj6zuRX12qBz7OoEf0ckTDini0BljFLA9JuF5v/pubhtml?gid=0&single=true