Keras is a good tool when getting started imo. But eventually it’s better to switch to PyTorch because it’s more pythonic and although tensorflow used to be the deployment standard PyTorch has caught up and add to that tf doesn’t even have a team working on it anymore iirc since everyone moved onto jax. I will say tho PyTorch can be very frustrating to work on initially because a lot of the internal optimizations that keras does are absent in PyTorch. I’ve never used PL tho.