Online LOOE-like for hyper-parameters selection.

[npinto]

Multiple (sgd_step_size0, l2_regularization) could be given and *fit() methods could use BLAS Level-3 operations when appropriate to allow for more data re-use and speed up the computation.

This is confusing... please consult me for more details ;-)

Loosely related work: Notes on Regularized Least-Squares by Rifkin. and. Lippert http://cbcl.mit.edu/publications/ps/MIT-CSAIL-TR-2007-025.pdf