This repository contains code that can be used to reproduce the results in our paper "Diversifying conformal selections".

Running simulations and experiments involving the Sharpe ratio and Markowitz objective involve using our dacs package, which contains custom projected gradient descent (PGD) solvers. Navigate to the divSelect directory and

• Install dacs:

  pip install -e . --verbose

• The install prerequisites are:

  mamba update -y conda mamba
  mamba env create -f environment.yml
  mamba activate dacs
  poetry install --no-root

sharpe-markowitz_sims_settings1-2 and sharpe-markowitz_sims_settings3-4 contain code to reproduce all of our simulation results involving the Sharpe ratio and Markowitz objective. The files sherlock_sharpe_driver.py and sherlock_markowitz_driver.py run our method, using our custom PGD solvers. The scripts sherlock_sharpe_driver_cvxpy.py or sherlock_markowitz_driver_cvxpy.py in sharpe-markowitz_sims_settings1-2 reproduce results using the MOSEK solver for reward computation. Additionally, the scripts optimized_sharpe_get_histogram.py and optimized_markowitz_get_histogram.py are used to construct the baseline distributions $\widehat{F}^{\varphi}_{\mathrm{baseline}}$ described in the paper.

The directory underrep_sims contains code for our underrepresentation index simulations. To reproduce our results, run either sherlock_cluster_driver.py or sherlock_cluster_driver_100.py (the latter uses $m=100$ test samples).

drug_exper contains code for our drug discovery experiments. The file data_preprocess.ipynb cleans the data, constructs the Tanimoto similarity matrix, and generates random train/calibration/test splits of the dataset. The file sherlock_train_dti.py constructs calibration and test scores, corresponding to each of these train-test splits, by training a small neural neural network on the training data. The files sherlock_sharpe_drug_driver.py and sherlock_markowitz_drug_driver.py run our method, using the custom PGD solvers, and drug_optimized_sharpe_results.py and drug_optimized_markowitz_results.py construct the baseline distributions $\widehat{F}^{\varphi}_{\mathrm{baseline}}$. The raw original dataset for these experiments is from: https://www.guidetopharmacology.org/DATA/gpcr_interactions.csv.

hiring_exper contains code for our job hiring experiments. All preprocessing code as well as code to run our method is found in the notebook hiring_exper.ipynb. The dataset for these experiments is from https://www.kaggle.com/datasets/benroshan/factors-affecting-campus-placement.