add method to compute average treatment effects #73
Conversation
|
build failures likely due to use of external packages in the new notebook. Should I add them as dev dependencies (or I can install them inside the notebook; unsure if that conforms to your style guide). Edit: done |
Implement ATE estimator by averaging pseudo outcome
|
@apoorvalal Thanks a lot for your kind words and effort! :) I'll get to your PR asap - hopefully on the weekend! |
|
A simple |
|
addressed most of your (great) suggestions in this commit. I think the conflicts are likely straightforward to resolve (amended a the drlearner test file, added the feature desc to changelog, and the usual pixi lock conflict |
* Move fixed propensity model to utils. * Add changelog entry. * Add tests. * Fix typo. * Fix class reference.
This reverts commit 3b2777c.
Minor adaptations to your PR
|
Ready to merge pending (opaque) pixi trouble (edit: and mypy decided to get mad at a different commit] |
There was a problem hiding this comment.
LGTM! Thanks a lot for your contribution @apoorvalal ! :)
I'll create a new release for version 0.9.0 soon.
Thanks for the nice work on this package; it addresses a lot of pain-points I have with
econML. An ATE estimator method was missing (possibly because that might not be the most obvious intended use case for the package, but the rest of the API / package ergonomics are nice enough that I decided to use it for that as well), so I added it (only todrlearner, since the variance-of-pseudooutcome construction for standard error is justified in that case but not for T-learner etc).I've also added a notebook in the
docs/examplesdirectory that verifies that estimates line up with what one gets fromeconMLanddoubleMLon a simulated dataset.Checklist
CHANGELOG.rstentry