In my project, I’m running training for multiple datasets each with multiple hyperparameters. The results aren’t really comparable from one dataset to the other, so I’d like to use either the workspace graphs or maybe the report feature (I haven’t tested it yet) to break down how the various hyperparameter combinations performed for each dataset.
For now, in my script, I’m using the name of the dataset as the run’s group name so I can group the runs by dataset in the WandB UI. I can get graphs to show aggregated data for each run group, but I can’t get them to show each run’s data for a single run group.
In other words, what I’m trying to do is to split runs based on their group name, instead of aggregating them. Is it even possible to do on WandB? I haven’t found how to do it yet. If not, what is the correct way to do it?
I know about sweeps but I’d rather not use them since I only use WandB for tracking runs and logging data, not for starting them and managing hyperparameters.