== FEATURES/PROBLEMS:

* Applying transforms to query vectors
* Allow objects to be passed in as transforms.
  * Hashes might be enough, but a faster data structure might be a good option.
* Detect the optimal dimension reduction in LSA.
  * This needs some benchmark. Low number of dimensions can be effective enough.
  * http://nlp.stanford.edu/IR-book/html/htmledition/latent-semantic-indexing-1.html
* Implement Probabilistic latent semantic analysis
* Implement Latent Dirichlet Allocation

* Matrix transformer has to popout the matrix of VectorSpace::Model and reassign it, get rid of this.