In Search of the Infamous Error Rate

by Godfried Toussaint

In questions relating to patterns,
It's finding the errors that matters.
Resubstitution,
A so-called solution,
Will oft yield a figure that flatters.

Well fix it; that's easy to say.
Obtaining new sample's the way.
But with data sets small,
You can't do much at all,
Only use half for training you may.

Our quest for a method efficient,
Has taxed even those called omnicient.
Leave-one-out seems the highest,
Gives a figure unbiased,
But the variance! Oh no! It's deficient.