Could multi_arm_causal_forest deal with ordered treatment arms?

[RenJump]

Recently I am trying to use the multi_arm_causal_forest to deal with a multi-treatment problem. Our experiment has one control and two treatments conditions. There is also a natural order associated with the two treatment conditions.

Just as your great work "Quasi-Oracle Estimation of Heterogeneous Treatment Effects" said, "When implementing variants of this approach in practice, different choices of Λn{τ (·)} may be needed to reflect relationships between the treatment effects of different arms, e.g. whether there is a natural ordering of treatment arms, or if there are some arms that we believe a priori to have similar effects." Could you advise on what kind of choices would be suitable in our case? Are there any related examples we could reference as we could not find any in the paper.

[swager]

That's a great question. That's a direction that would be natural to explore; however, we haven't (yet) built anything like this in GRF.