you need applications and queryable storage for maintaining models both for production use, internal testing/validation/research, and everything in between. Mai

you need applications and queryable storage for maintaining models both for production use, internal testing/validation/research, and everything in between. Maintaining a model - or collection of models - implies maintaining a ton of data that can ideally be parameterized (learning data, build info, tuning parameters, purpose, priority, etc) and there are a lot of sophisticated ways to handle this all efficiently mostly dictated by your problem space (algorithm, data volume, response time, research methodology, etc). This is all outside (or inside) of whatever application runs your actual business model and the infrastructure necessary to query your "live" model(s) in a distributed/replicated environment if necessary (as is for most big data projects).