Hyperparameter tuning do be like that.

It’s called “trial and error” and it’s a valid scientific method. It sucks though if you’re the first one to publish it in your thesis. Examiners like to see some citations for your methods and saying you just “guessed” will get you some grief.