METRIK: Measurement-EfficienT Randomized Controlled Trials using Transformers with Input MasKing

METRIK is a tool to reduce the number of measurements acquired in a clinical RCT.

Table of Contents

• Environment Setup
• Dataset request
• Run METRIK
• Developer
• Cite this work
• License

Environment setup

See packages are listed under env.yml (you may face installation issues when creating environment from this file directly).

Dataset Request

To obtain the datasets used in the paper (see details in paper), submit requests to [NINDS] (https://www.ninds.nih.gov/current-research/research-funded-ninds/clinical-research/archived-clinical-research-datasets).

Download these into the datasets directory.

Run METRIK

From the slurm directory, run the following commands. This will launch jobs using slurm to run METRIK. See comments (TODO) within files to adapt script for your purposes.

First, train the initial imputation models.

./slurm/rct_train_mask_0.sh

Next, generate candidate PMD-imputer models.

./slurm/rct_train_mask_1.sh

Aferwards, run the analysis script to find solutions.

./slurm/analysis.sh

Developer

Sayeri Lala. For any questions, comments or suggestions, please reach me at slala@princeton.edu.

Code for the [differentiable masking layer] (https://arxiv.org/pdf/2010.02066) was adapted from the [link] (https://github.com/RobertCsordas/modules).

Code for the [MVTS algorithm] (https://dl.acm.org/doi/10.1145/3447548.3467401) was adapted from this [link] (https://github.com/gzerveas/mvts_transformer).

Cite this work

Cite our work using the following bitex entry:

@article{lala2024metrik,
  title={METRIK: Measurement-Efficient Randomized Controlled Trials using Transformers with Input Masking},
  author={Lala, Sayeri and Jha, Niraj K},
  journal={arXiv preprint arXiv:2406.16351},
  year={2024}
}

License

Copyright (c) 2024, Sayeri Lala, Jha Lab, and The Trustees of Princeton University. All rights reserved.

See License file for more details.