remove dependency on NFS

[andaag]

Hi

I stumbled over your "Advanced Machine Learning with scikit-learn" on youtube (which was really good), and found this project. It's exactly what I was looking for. However I don't have NFS set up (and with my ipcluster engines it wont be easy to set up).

I modified persist_cv_splits to essentially do:
[client[engine_id].apply(distributed_joblib_dump, cv_fold, cv_split_filename) for engine_id in one_engine_per_host]

And it works very well, however my file structure is the same for all hosts, so I can "hardcode" that part. Maybe it'd be better to use relative paths?

I can make a tiny pull request for this of course, but I wanted your input first. I hadn't used ipython parallel for clustering before, so I'm not sure if people generally setup one project directory (in which case using relative paths would be fine), or if they tend to be in $HOME for example.

[ogrisel]

Hi sorry I had not seen this message. This would indeed be an interesting pull request. Please add a test to no ship the data to the engines if a file with the same name is already there, unless you explicitly pass an overwrite=True argument.