Bauxitiego/

uplift-modeling-synthetic-benchmark

100,000 training rows, 20,000 validation rows, generated for uplift-modeling's Gate 0: checking that meta-learners (S/T/X-learner) actually recover a real treatment effect before trusting them on data where no individual ground truth is ever available - which is true of essentially all real causal-inference data, by the fundamental problem of causal inference (nobody observes both potential outcomes for the same… See the full description on the dataset page: https://huggingface.co/datasets/Bauxitiego/uplift-modeling-synthetic-benchmark.