A probabilistic model for data cube compression and query approximation