normalizing density estimates instead of targeting qn
nhejazi opened this issue · 1 comments
nhejazi commented
In order to ensure efficiency of the IPCW-TMLE, we are presented with two avenues by which we may proceed (1) we could add an extra targeting step, wherein we obtain better estimates of q_n (which itself corresponds to only the part of the EIF having to do with the baseline covariates W) and a corresponding updated value of the fluctuation nuisance parameter \epsilon; or, (2) it is apparently just as good to normalize the finalized IPCW density estimates such that the estimated values form a proper density (according to vdL). Obviously, the latter of these is easier.