have you compared gainst #5491?

You can't really add a dependency to scikit-learn, but you could add this to the scikit-learn-contrib organization https://github.com/scikit-learn-contrib/

Hi Andreas,

Is it a policy for scikit-learn not to have a dependency? This tool is core
to what a machine learnist does and so it would be really valuable to have
this in the core module - don't you agree?

Have a look at the introduction:
http://optunity.readthedocs.org/en/latest/index.html. Its really cool what
it does

@joshlk adding dependencies to scikit-learn is a pretty big addition, and not one that should without careful thought and planning. scikit-learn is great in that it only requires scipy and numpy, because it uses those functions extensively across the codebase; i don't think it would be worth it to add an extra dependency just for one feature, especially when it seems something similar is already being implemented without said dependency.

@joshlk how is what it does different from the PR I linked to?

actually optunity doesn't even implement the optimization algorithms itself. It only implements particle swarm, the rest are thin wrappers. So you need hyperopt and bayesopt as dependencies as well.

@joshlk, I think it's great that you've produced this wrapper. I also know that it's not going to get merged into scikit-learn, without actually reimplementing or copying the particle swarm optimisation into Python, with good evidence that this is a worthwhile thing for us to provide and maintain as part of the main package.

I strongly recommend setting up a scikit-learn-contrib project that does this and having it referenced from our related work section, because I think wrappers for good ML projects without scikit-learn-compatible interfaces are great stuff to have available for our users.

As such, I'm going to close this PR, not because I don't respect your work, but because this will not be merged into the core scikit-learn codebase.

The question does come to mind, @amueller, of whether we should be listing/cross-referencing related projects in relevant sections of the narrative documentation, so as to say that there are external alternatives.