Rigorous LASSO and causal inference
GalAmedi opened this issue · comments
Hi.
I have two requests for future development:
-
I want to estimate causal LASSO on large datasets. Currently I do it with the hdm package. I think it internally uses glmnet, thus it quickly becomes infeasible as I increase the size of the data set. Could you incorporate a standardized procedure to estimate this?
-
In order to quickly manually implement a causal LASSO one can run a rigorous LASSO of X separately on Y and D, and then use the partialled-out results to estimate an average treatment effect. There is currently no option to estimate a rigorous LASSO in the package, could it be implemented?
Gal
First of all, I have no idea what a "rigorous LASSO" is. Second, it sounds like you're asking: Can we rewrite the entire hdm package so that it uses biglasso internally instead of glmnet? This may indeed be a worthwhile project, but I don't see how it has anything to do with this package. It's possible I'm not understanding your request, but it sounds more like an hdm issue than a biglasso issue to me.