What importance weights correct covariate shift?

Assume covariate shift: the source and target domains have different input distributions, but Pₛ(y|x)=Pₜ(y|x). Which sample weight makes source risk resemble target risk?