As for those model comparisons (bias/RMSE), I assume these stats are based on the training set? If so, than I would rather see the RMSE stats on a hold-out test set as a fair comparison for robustness.
The RMSE was with respect to estimating a marginal causal effect across thousands of datasets. It had nothing to do with the accuracy of the outcome predictions. For details on the 2019 ACIC Data Challenge see sites.google.com/view/acic2019datachallenge/home
this is a great presentation - thanks for uploading!
39:10 for list of references
As for those model comparisons (bias/RMSE), I assume these stats are based on the training set? If so, than I would rather see the RMSE stats on a hold-out test set as a fair comparison for robustness.
The RMSE was with respect to estimating a marginal causal effect across thousands of datasets. It had nothing to do with the accuracy of the outcome predictions. For details on the 2019 ACIC Data Challenge see sites.google.com/view/acic2019datachallenge/home
@@putnamdatasciences1309 Thank you for providing this context. I like the thought process behind this ML algo, so I will definitely run some tests.