cmrdesign Python Package

Python implementation of cmrdesign, the applied Conditional Minimax Regret design-rule package.

The methods accompany When and How to Pilot: Design Rules for Two-Wave Experiments by Juan C. Yamin. This package is implementation-focused and uses simulated examples only; paper replications and empirical simulations are intentionally kept out of the package repository.

Installation

Install the Python package from PyPI:

python -m pip install cmrdesign

For exact reproducibility, pin the release version:

python -m pip install cmrdesign==0.1.0

For the development version from GitHub:

python -m pip install "cmrdesign @ git+https://github.com/juancyamin/cmrdesign.git#subdirectory=python"

For local development from the repository root:

python -m pip install -e python

Quick Start

import numpy as np
import cmrdesign as cmr

rng = np.random.default_rng(123)
d = np.r_[np.ones(40), np.zeros(40)]
y = np.r_[rng.beta(2, 5, 40), rng.beta(4, 4, 40)]

fit = cmr.cmr_two_arm(y, d, alpha=0.05, method="auto")
print(fit.pi)
print(fit.U_CMR)

allocation = cmr.realize_allocation(fit, n_main=1000)
print(allocation.counts)
print(allocation.realized_U_CMR)

Implemented Surface

• cmr_two_arm() and cmr_binary().
• method="auto" for the applied bounded/binary default.
• Maurer–Pontil bounded-outcome variance rectangles.
• Exact Bernoulli folded-binomial variance rectangles.
• Martinez-Taboada–Ramdas (MTR) bounded-outcome rectangles.
• Two-arm unbounded-outcome median-of-means rectangles via cmr_unbounded().
• Shared-control multi-arm CMR via cmr_multiarm().
• Stratified CMR via cmr_stratified().
• Multiple-outcome CMR for weighted-index and co-primary workflows via cmr_multiple_outcomes().
• Proxy/delayed-outcome CMR bridge widening via cmr_proxy().
• Pilot-planning helpers from Appendix E of the accompanying paper (Yamin 2026) via cmr_plan().
• Integer main-wave counts from CMR shares via realize_allocation().
• Expert rectangle helpers via rectangle_*() and cmr_*_from_rectangle().

See the repository-level docs for applied guidance:

• Quick Start
• Choosing a Method
• Methods
• Pilot Planning

From the repository root, run local checks with:

python -m pip install -e python
python -m unittest discover -s python/tests -v