You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
First: for linear models it most definitely will be a projection of the cate and not well-specifided.
Second: the current version cannot handle opt_rewieghted and will throw an error. The reason being that StatsModelsLinearRegression can only handle integer weights when doing inference with sample weights and treats these weights as composite observations. However, the weights created by opt_reweighted have a different interpretation. One way to handle these opt_reweighted would be to re-scale the outcome and the features based on the weights and call the statsmodels if we want to enable this flag. But given "first" not clear we want to do this.
The text was updated successfully, but these errors were encountered:
We should disable the opt_reweighted here and hard code it to False.
https://github.com/microsoft/EconML/blob/ffdfda8bf733313fde4f5f762b6fa3c62d7e1531/econml/ortho_iv.py#L1412
First: for linear models it most definitely will be a projection of the cate and not well-specifided.
Second: the current version cannot handle opt_rewieghted and will throw an error. The reason being that StatsModelsLinearRegression can only handle integer weights when doing inference with sample weights and treats these weights as composite observations. However, the weights created by opt_reweighted have a different interpretation. One way to handle these opt_reweighted would be to re-scale the outcome and the features based on the weights and call the statsmodels if we want to enable this flag. But given "first" not clear we want to do this.
The text was updated successfully, but these errors were encountered: