Prediction with eventual almost sure guarantees.

We study the problem of predicting the properties of a probabilistic model and the next outcome of the model sequentially in the infinite horizon, so that the perdition will make finitely many errors with probability 1. We introduce a general framework that models such predication problems. We prove some general properties of the framework, and show some concrete examples with the application of such properties.