How to Train a Detecter on Buildings

OpenStreetMap is a huge resource and I'm just thinking about how can I quickly get if we're talking about building classification for example. Can you talk a little bit more about the training process because I think we understand now that we can go in and we can select these objects manually or we can use a no on data source to make that selection. What does the training process look like because I'm imagining anyway that it's unlikely I'm going to get it right the very first time.