This is a set of simple demos of Storm indicating how it can be used for aggregation of events and for Bayesian on-line learning
See SnappedCounterTest for a beginning of a test for the counter. This test uses EventSpout to create a stream of keys and values, SnappedCounter to count them and FileBolt to persist them.
Note the use of tuple acking to avoid any sort of retry log in the counter.
The BanditTrainer shows how a two-armed bandit can be solved using a model that I call the beta-Bayesian model.
The BetaWalk implements a random walk that has assymptotic beta distribution. This is useful for modeling conversion probabilities that vary in time but which have realistic distribution.