causaltune
causaltune copied to clipboard
Test multivalued categorical treatment dataset
Extend pipeline to accept multi-valued categorical treatment datasets.
Checklist
- [x] Test single multi-valued dataset
- [x] Create notebook for Multi-Valued Treatment examples
- [x] Prepare minimum 3 datasets for comparison
maybe this one? https://github.com/AMLab-Amsterdam/CEVAE/blob/master/datasets/TWINS/ReadmeTwins Looks like it has two outcome variables, denoting the death of each twin (I know.. quite dark)
ah nevermind, that one has a continuous treatment variable..
Here's a list of multivariate datasets (table 1) https://arxiv.org/pdf/1702.06385.pdf @Ed-Zh would you have time to do some research on these datasets and check if they are publicly avaliable? 'm happy to write the wrappers
@Ed-Zh maybe also have a look at this one https://cran.r-project.org/web/packages/causaldata/causaldata.pdf There's a link to a github repo with all datasets. Some seem to have categorical outcomes
Additional Data sources:
-
Patient Survival One year after treatment. Description here • Data source
-
bdsianesi - Effect of education on earnings (UK) Description here • Data source
-
close_college - Effect of college education on earnings (USA, 1995) Description here - pg 10 • Data source
Some sources for causal model benchmarking/data (@markharley):
- https://github.com/IBM-HRL-MLHLS/IBM-Causal-Inference-Benchmarking-Framework
- https://acic2022.mathematica.org/
- https://github.com/rguo12/awesome-causality-data
Hi I'm wondering if multi-value categorical treatments has been supported or not. If not should I open another request for this feature? I see from the readme that it can be done in two weeks if requested, is it currently under development?
Thank you.
@peterzhb thanks for your interest in our package! They are not supported yet - are there specific EconML models you care about? Just so we know what to focus our testing on.
@peterzhb We have recently merged support for multivalue treatments into our main branch, do you want to give it a try? Unfortunately it required changes in dowhy as well, those have been merged into their main branch but not yet packaged to pypi afaik, so you'd have to run dowhy off the tag in our requirements.txt until they do