How do I “choose the best action” without having to specify treatments?

[tonyabracadabra]

In the real world application, people tend to have their expected goal and seeking for approaches to achieve that goal (given a model, then conduct counterfactual generation), is this within the scope of project azua? I think azua does an amazing job in proposing an end to end framework for conducting CATE but we also need to know which variables we should intervene upon in order to calculate the CATE. If I have potentially many treatments, does that mean I need to run through all the treatment combinations in order to find the one that maximizes the CATE?

What do you guys think? I thought to better achieving the goal for “choose the best action” we may want to have some interface like the branch of work that DiCE did by generating a series of counterfactual recourses, or there may be opportunities to propose a even more unified model for doing this counterfactual CATE generation?