I am working on multiple deep learning projects, most of them in the area of computer vision. For many of them I create multiple models, try different approaches, use various model architectures. And of course I try to optimize hyperparameters for each model.
Now, that itself works fine. However, I start to lose track of all the various parameters and model layouts I tried. The problem is, sometimes for example I want to re-train a model from a past project with a new data set, but using the same hyperparameters from the last (best) successful training. So I need to look up that project's documentation, or I have some hyperparameters saved in a text or Excel file, etc.pp.
For me that feels a bit cumbersome. I bet I am not the only one facing this problem, surely there must be a better way than "remembering" all the hyperparamters from all projects / models manually via text files and alike.
What are your experiences, have you found a better software / solution / approach / best practice / workflow for that? I must admit, I would welcome a software to aid with that a lot.