Documentation for configuration options and dataset registration

[Sann5]

📚 Describe the documentation issue

I would like to use GraphGym to find good models for a suite of private datasets/tasks. Let us assume that I intend to follow the first three steps in the documentation on how to run a batch of experiments. To do so I would need to specify two config files and a grid file.

• Issue No.1: How can I customize these config and grid files? I believe users would greatly benefit from documentation explaining and specifying all possible fields in the config/grids files, as well as all valid options for each field. Simply providing example grid and config files leaves the user unsure of 1) the options they have and 2) what exact models they are deciding on since the field explanations are missing.
• Issue No.2: How do I register my datasets? From what I have figured out one must create a costume dataloader.py analogous to example.py. But then what? Should I run this script from the terminal before I run configs_gen.py or does my /graphgym/custom_graphgym/loader/dataloader.pyget called by it? Should the config and grid files I mentioned above already have the name of one of the datasets that I intend to use?

I appreciate your support and attention in advance. Cheers!

Suggest a potential alternative/fix

No response

[rusty1s]

I definitely agree that there exists parts of missing documentation in GraphGym, really sorry about that. We are trying to improve on our documentation efforts in the upcoming releases.

1. We could also think about documenting this in the configs directly as well, e.g., here.
2. Yes, that is correct. There is no need to run everything, the custom modules/functions will be imported in the graphgym namespace automatically and can be referenced in the config. For example, you can load datasets defined here via

   format: PyG
   name: QM7b