Add an API for instrumental variable approach and doubly-robust approach

[adam2392]

trafficstars

Currently, working on a DML meta-estimator that we can plug in any arbitrary trees. This could probably replace the DML set of classes which are pretty confusing (to me).

Another set of API we would want to complete in order to achieve feature parity is:

1. instrumental variables
2. doubly-robust learning approach

The IV approach is just a two-stage regression that requires explicitly passing in instrumental variable array. It is more or less pretty similar to the DML in terms of API and code.

The Doubly-robust (DR) approach is very similar conceptually to the DML. I"m still figuring out exactly what are the detailed differences. I think DR just uses some form of augmented-IPW to estimate the nuisance model for the outcomes, rather than your standard regression.

[adam2392]

Possibly relevant papers:

• https://proceedings.mlr.press/v70/hartford17a/hartford17a.pdf
• https://eml.berkeley.edu/~powell/npiv.pdf

[adam2392]

Note the DML API is probably highly relevant. Xref: #52