cleanup clustering

[marscher]

• remove mdtraj reference

[marscher]

Since the minRMSD metric is contained within the code branches of kmeans and regspace, we need to dynamically load the definition of the distance function from PyEMMA (which will then build the mdtraj dependent code). PyEMMA needs some kind of library dir exposure like mdtraj already has (eg. mdtraj.capi()).

However this approach seems very fragmented and very hard to understand.

One could also think of adding a so called setuptools "Feature" to opt-in mdtraj as a dependency. If the users are going to use the lib with pyemma they most likely have/want mdtraj too...
This would just add the need for some conditional compilation, which is easier to achieve than dynamic loading.

@clonker what do you think?

[marscher]

@clonker During linkage of the dynamic linking stuff on Windows, the linker complains about the missing init function of minRMSD_metric (which is true, we just pretended it to be a python module). This makes me think that we could avoid all the dynamic linkage hassle by making it a proper module, which we import in regspatial and kmeans.c to expose the functionality. This would keep the code more portable.

[clonker]

yes i agree, that does sound more reasonable and dynamic linking should be avoided whenever possible (imo)