whether the cross-validation (CV) step is mandatory ?

[zhanglin-alt]

Dear Prof. Tadano,

I would like to inquire whether the cross-validation (CV) step is mandatory. I've observed that CV computation can be time-consuming, and under standard conditions, a tolerance of 1e-6 appears sufficiently convergent for fitting errors. Consequently, if I set the L1_ALPHA parameter as low as possible—for instance, 1e-10—could this allow me to bypass the CV validation step?

Best regards！

[ttadano]

It is not recommended to use a specific L1_ALPHA value without validation. If the number of training data sets is small, a small L1_ALPHA value can result in overfitting.

[zhanglin-alt]

Thank you for your insightful advice.