When working on an opensource project, you'll become an expert on the few functionalities (and code that realises it) of that single project that you will be working on - while you'll be oblivious about everything else in the subfield. IMO just doing a project that's typical for the subfield will allow you to learn much more. So, for example, when learning ML, don't go do obscure work on R codebase, but instead try some simple challenges from Kaggle.