Is it possible tune the the pipeline as a whole? i.e., find the best version of processing and training parameters (combined) for given input?