[yetanotherphd]

That looks right to me. This is a very important calculation to do in order to avoid getting 100x the number of samples that you really need.

Essentially the same calculation applies to classification error.

[ska]

The real (and all to common to run into ) problem isn't having 100x the samples you really need, it is having 1/10 the samples you really need... And not knowing that.