erikbern/uncertainty

Blog question

BruceBinBoxing opened this issue · 1 comments

BruceBinBoxing commented

Hi Erikbern,
Thanks for your blog. Could you deeply explain why we need 10000 examples for bellow A/B test?

For instance, this is useful when you’re running an A/B test – if you want to detect a 1% difference then you need something on the order of 0.01e-2 = 10000 samples.

erikbern commented

0.01^-2

does that make sense?