* a few tweaks to speed up data generation * del variable to save memory * switch to random numpy arrays
* Refactor linear modelling and add new coordinate descent updater * Allow unsorted column iterator * Add prediction cacheing to gblinear