Adaptive approximate querying of large sparse binary data sets via probabilistic model averaging