next-to-naive hit and miss sampling
if we can model our weight distribution somehow...
- generate training samples and build simple model distribution with as small as possible
- generate event using g
- calculate weight
- generate random number
- return and if