From: Brian Schrom
It would be desirable that the machine learning topic area be designed a bit more comprehensively than a single library. For example, I should be able to run a GA on the same data as the DE. I would hope that they have the same API. This somewhat implies to me that DE/GA are instances of training algorithms, for which, many are possible.
There are many things that go into a ML solution. Data conditioning, model training, model validation, model evaluation. Training in batch modes, instance based, etc. Just to name a couple issues...
I think that numerical optimization is more general than its application to machine learning. While you argue that you'd like to see the library go deep in machine learning I don't see anything wrong with a general purpose numerical optimization library. However, DE is more general than numerical optimization, while the library seems to be an application of DE to numerical optimization. I'm not sure that DE is even one of the best ways to do numerical optimization (I'm pretty sure genetic algorithm is a bad way, so I'm skeptical), and I know I'd like a more general interface to a DE engine than one designed for numerical optimization problems. I'd like to see the library go wide for max applicability rather than deep. Hopefully that will keep it out of all the different cans of worms that could be opened in each application domain. Put numerical optimization and machine learning applications in the example code of the library and submit just the core engine as the library itself. Narrowing the scope of a library submission is usually the path to success. Regards, Luke