[MRG+1] added a new section on model persistence

[raulgarreta]

Fix for issue #1332

[ogrisel]

...which is more efficient on objects that carry large numpy arrays internally as is often the case for fitted scikit-learn estimators.

[ogrisel]

Apart from my comments, this looks good. +1 for merge once they are addressed.

[raulgarreta]

Addressed ogrisel comments in 640bff9

[ogrisel]

Ping @mblondel @amueller @GaelVaroquaux

[coveralls]

Coverage increased (+0.01%) when pulling 640bff9 on raulgarreta:model_persistence_doc into b70a481 on scikit-learn:master.

[jnothman]

I'm not sure about this positioning. For example, after (or within) model selection and evaluation seems more coherent.

[mblondel]

+1

[mblondel]

For the sake of consistency, I'd rather avoid the use of "you": we don't address the reader directly in the rest of the doc. Besides that +1 for merge once @jnothman 's comments have been taken into account.

[jnothman]

Fixed by #3317