Doing machine learning in production is very different then for proof of concepts or in education. \nOne of the hardest parts is keeping models updated.