BrainDump is sometimes too computationally expensive on some large applications. Developers may want to only enable part of the pipeline, e.g. if one has a huge feature table, he may not want to do a full count over it. Some users are not interested in good-turing estimator, etc.
We should be able to add switches to each part of the pipeline.