Proposal: Integrating CATENets into EconML

[AlxndrMlk]

Hi,

CATENets (https://github.com/AliciaCurth/CATENets) is a Python library that implements a number of flexible neural-network-based architectures for conditional average treatment effect estimation written by Alicia Curth and collagues (back then at Cambridge).

The library implements classic algorithms like TARNet (Shalit et al., 2017; https://proceedings.mlr.press/v70/shalit17a/shalit17a.pdf) and more modern variations like DragonNet (Shi et al., 2019; https://arxiv.org/pdf/1906.02120)

CATENets has a user-friendly interface, that could be relatively easily integrated with EconML's API.

I believe that CATENets could provide a nice extension of EconML capabilities.

The library requires either PyTorch or JAX as a backend,

What are your thoughts on this idea?