Double checking correctness of default behavior of propensity.func() when NULL about personalized HOT 1 CLOSED

This is fine. Note how in the weighting method you divide each term by $T_i\pi(x_i) + (1 - T_i) / 2$, where $T_i \in {1, -1}$. This is just $Pr(T = T_i | X = x_i)$. The idea is the same for multiple treatments, you are dividing by the probability of receiving the treatment you actually received. In the multiple treatment case I allow the user to specify a vector of probabilities (which must be the probability of receiving what they received) or a matrix of probabilities with each column being the probability of receiving a particular treatment, which i then just convert to the first thing

from personalized.