[nraynaud]

yeah, this one is quite intuitive, but it reduces the training sample size.

[Houshalter]

Once you find the optimal parameters, you can then train the model again on the entire dataset.