Hi:

When collecting a new set of samples for validation is expensive, one alternative is to use the k-fold validation to mimic the situation in that we have k sets of samples for validation.

However, we have to make sure these k units have identical characteristics. For example, if we have the numbers of normal patients and diseased ones to be 2:3 in the original set of samples, we ought to maintain this ratio for each of the k units.
Suppose the original set has two types of patients 2*n1 and 3*n2. We have to choose the k to be a common factor of n1 and n2 so that the two types of patients in each of the k units are 2*m1 and 3*m2, where n1 = k*m1 and n2 = k*m2.

What if both n1 and n2 are prime numbers? More generally, what if n1 and n2 do not have a common factor?

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Cosine wrote:

Hi:

When collecting a new set of samples for validation is expensive,
one alternative is to use the k-fold validation to mimic the
situation in that we have k sets of samples for validation.

However, we have to make sure these k units have identical
characteristics. For example, if we have the numbers of normal
patients and diseased ones to be 2:3 in the original set of samples,
we ought to maintain this ratio for each of the k units. Suppose the
original set has two types of patients 2*n1 and 3*n2. We have to
choose the k to be a common factor of n1 and n2 so that the two types
of patients in each of the k units are 2*m1 and 3*m2, where n1 = k*m1
and n2 = k*m2.

What if both n1 and n2 are prime numbers? More generally, what if
n1 and n2 do not have a common factor?

THe obvious solution here is to do the analyses in terms of quantities
whose interpretation is not sample-size dependent.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)