From: Joaquim Almgren Gāndara (firstname.lastname@example.org)
Date: Sun Dec 03 2000 - 11:16:03 MST
> Ultimately one wants a model with as many free parameters as there are
> "implicit in the data." But this is not always known, or even well-defined!
And if you have too many free parameters, you get overfitting.
> So in this case, which is a real-life example, sophisticated generalization
> methods provide more OPPORTUNITY for overfitting, but also provide
> the opportunity to avoid overfitting by doing the sophisticated analysis
OK, but that sounds like a bit of a hybrid to me. You're talking about
generalisation *and* analysis. I was referring to generalisation purely as a way
of classifying roughly, with some acceptable error, e.g. "apples are edible". In
addition, I'm assuming that the generalisation is at a subconscious level, or a
mechanism that the mind employs more or less implicitly. Your generalisation
methods sound more conscious and explicit.
- Joaquim Gāndara
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:35 MDT