This pull request proposes adding a new method for estimating inverse probability weights for IPW and TMLE methods via empirical loss-based estimation of Riesz Representers.
The code for performing the Riesz Representer estimation has been abstracted to a separate package, SuperRiesz ( https://github.com/herbps10/SuperRiesz), which is intended to be in the style of mlr3superlearner.
The main difference between the Riesz Representer approach and estimating probability of treatment weights is that the probability of treatment weights need to be cumulatively multiplied to form the final weights, while the Riesz Representers do not. Therefore there is some extra code added to lmtp so that whether the weights need to be cumulated or not can be tracked.
This pull request proposes adding a new method for estimating inverse probability weights for IPW and TMLE methods via empirical loss-based estimation of Riesz Representers.
The code for performing the Riesz Representer estimation has been abstracted to a separate package, SuperRiesz ( https://github.com/herbps10/SuperRiesz), which is intended to be in the style of mlr3superlearner.
The main difference between the Riesz Representer approach and estimating probability of treatment weights is that the probability of treatment weights need to be cumulatively multiplied to form the final weights, while the Riesz Representers do not. Therefore there is some extra code added to
lmtpso that whether the weights need to be cumulated or not can be tracked.