Learning stateful models for network honeypots