Sign inSign up

pageman/splitup-reproducibility

By pageman

•Updated 8 months ago

SplitUP estimator (arXiv:2601.15254) repro pkg see also DOI: 10.13140/RG.2.2.20833.77925

Image
Machine learning & AI
0

1.0K

pageman/splitup-reproducibility repository overview

⁠SplitUP Reproducibility Package

Monte Carlo simulations for the SplitUP estimator from:

"Many Experiments, Few Repetitions, Unpaired Data, and Sparse Effects: Is Causal Inference Possible?" (arXiv:2601.15254)

⁠What is SplitUP?

SplitUP is a novel estimator for causal inference when treatment (X) and outcome (Y) are never jointly observed - a common scenario in genomics, economics, and medical research where data linkage is
impossible due to privacy or technical constraints.

⁠Estimators Implemented

EstimatorDescription
NaiveOLSBaseline ordinary least squares (biased)
TS-IVTwo-stage instrumental variable estimator
UP-GMMUnpaired generalized method of moments
SplitUPCross-fold bias-corrected estimator (proposed)

⁠Quick Start

docker pull pageman/splitup-reproducibility:latest                                                                                                                                                            
docker run -p 8888:8888 pageman/splitup-reproducibility:latest                                                                                                                                                
                                                                                                                                                                                                              
Then open http://localhost:8888 in your browser to access Jupyter Lab.                                                                                                                                        
                                                                                                                                                                                                              
Key Features                                                                                                                                                                                                  
                                                                                                                                                                                                              
- Full Python implementation of all estimators from the paper                                                                                                                                                 
- Configurable simulation parameters (m, n, r, d, sparsity)                                                                                                                                                   
- Reproducible Monte Carlo experiments                                                                                                                                                                        
- Visualization of estimator performance                                                                                                                                                                      
                                                                                                                                                                                                              
Environment                                                                                                                                                                                                   
                                                                                                                                                                                                              
- Python 3.12.2                                                                                                                                                                                               
- PyTorch 2.9.1                                                                                                                                                                                               
- NumPy 1.26.4                                                                                                                                                                                                
- SciPy 1.16.2                                                                                                                                                                                                
- Scikit-learn 1.7.2                                                                                                                                                                                          
                                                                                                                                                                                                              
Tags                                                                                                                                                                                                          
                                                                                                                                                                                                              
- latest - Most recent build                                                                                                                                                                                  
- 1.1.0 - Stable release                                                                                                                                                                                      
                                                                                                                                                                                                              
Citations                                                                                                                                                                                                     
                                                                                                                                                                                                              
@misc{schur2026experimentsrepetitionsunpaireddata,                                                                                                                                                            
  title={Many Experiments, Few Repetitions, Unpaired Data, and Sparse Effects: Is Causal Inference Possible?},                                                                                                
  author={Felix Schur and Niklas Pfister and Peng Ding and Sach Mukherjee and Jonas Peters},                                                                                                                  
  year={2026},                                                                                                                                                                                                
  eprint={2601.15254},                                                                                                                                                                                        
  archivePrefix={arXiv},                                                                                                                                                                                      
  primaryClass={stat.ML},                                                                                                                                                                                     
  url={https://arxiv.org/abs/2601.15254}                                                                                                                                                                      
}                                                                                                                                                                                                             
                                                                                                                                                                                                              
@article{pajo2026splitup,                                                                                                                                                                                     
  author={Pajo, Paul},                                                                                                                                                                                        
  title={Finite-Sample Performance of SplitUP in Many-Environments Unpaired Instrumental Variables: A Simulation Study with Pilot Results},                                                                   
  month={January},                                                                                                                                                                                            
  year={2026},                                                                                                                                                                                                
  doi={10.13140/RG.2.2.20833.77925}                                                                                                                                                                           
}                                                                                                                                                                                                             
                                                                                                                                                                                                              
License                                                                                                                                                                                                       
                                                                                                                                                                                                              
MIT License - Free for academic and research use.                                                                                                                                                             

Tag summary

Content type

Image

Digest

sha256:15c919024…

Size

447.2 MB

Last updated

8 months ago

docker pull pageman/splitup-reproducibility