We have a lot of studies to do with importance sampling. For now, we'll stick to the behavior of the one-sided z-test.
Questions to be answered:
[ ] (From Slack thread) Importance sampling is good for generating rare events:
[ ] Is k-means or any adaptive way to localize the theta_k to large rejection areas necessary?
[ ] Can we simply make points in the alternative space without trying to localize? What's the penalty of just blindly putting points far into the alternative space?
We have a lot of studies to do with importance sampling. For now, we'll stick to the behavior of the one-sided z-test.
Questions to be answered: