GPyTorch implementation of Counterfactual Learning with Multioutput Deep Kernels, from Caron et. al. (2022). The folders contain all necessary code to replicate examples and simulated studies in the original paper.
The code implements Counterfactual Multitask Learning via Gaussian Processes Alaa & Van Der Schaar (2017) or via Deep Kernel Learning Caron et. al. (2022), based on scalable package implementation in Torch.
DKL is particularly useful in policy learning (RL) settings where the action space is large (and possibly combinatorial), in addition to large input/context space and multiple outcomes.