The correct CV bandwidth selection procedure incorporating weights

NSAPH-Software / CausalGPS

Matching on generalized propensity scores with continuous exposures

https://NSAPH-Software.github.io/CausalGPS/

24 stars 5 forks source link

The correct CV bandwidth selection procedure incorporating weights #197

Closed wxwx1993 closed 1 year ago

wxwx1993 commented 1 year ago

The cross-validation on counter_weights data needs to take weights into account. @jennyjyounglee could you also review the following changes?

Naeemkh commented 1 year ago

Thanks, @wxwx1993. Please update the NEWS.md with the changes.

Naeemkh commented 1 year ago

Hi @jennyjyounglee could you also review the submitted changes.

jennyjyounglee commented 1 year ago

Hi @wxwx1993, I tested again the bandwidth selection of locpol with the updated code. However, now even in the single-level data, the optimal bandwidth selected using KernSmooth and locpol is different where kernel smoothing working better...

Naeemkh commented 1 year ago

Hi @wxwx1993 and @jennyjyounglee. Should we merge this PR into the codebase?

Naeemkh commented 1 year ago

Rejecting this PR at this time. If you need to add your changes to the codebase, please:

Make sure that you start your branch on top of the develop branch.
Update NEWS.md file.
Make sure that all tests pass successfully.

Please let me know if you have any questions or need help.