PyOptEx

Package
Meta

PyOptEx (or Python Optimal Experiments) is a package designed to create optimal design of experiments with Python. It is fully open source and can be used for any purpose.

The package is designed for both engineers, and design of experiment researchers. Engineers can use the precreated functions to generate designs for their problems, especially the cost-optimal algorithms. Researchers can easily develop new metrics (criteria) and test them.

To generate experimental designs, there are two main options:

• Fixed structure: These designs have a fixed number of runs and fixed randomization structure, known upfront. Well-known designs include split-plot, strip-plot, and regular staggered-level designs. A specialization is also included for splitk-plot designs using the update formulas as described in Born and Goos (2025).

• Cost-optimal designs: These design generation algorithms follow a new DoE philosophy. Instead of fixing the number of runs and randomization structure, the algorithm optimizes directly based on the underlying resource constraints. The user must only specify a budget and a function which computes the resource consumption of a design. Go to Creating a cost-optimal design for an example. The currently implemented algorithm is CODEX.

NOTE: This package does not have a release version yet and is still under active development.

Main features

• The first complete Python package for optimal design of experiments. Model everything including continuous factors, categorical factors, mixtures, blocked experiments, split-plot experiments, staggered-level experiments.

• Intuitive design of experiments with cost-optimal designs for everyone. No longer requires expert statistical knowledge before creating experiments.

• Accounts for any constraint you require. Not only can you choose the randomization structure manually, or let the cost-optimal design algorithms figure it out automatically, you can also specify the physically possible factor combinations for a run.

• Augmenting designs was never easier. Simply read your initial design to a pandas dataframe and augment it by passing it as a prior.

• Customize any part of the algorithm, including the optimization criteria (metrics), linear model, encoding of the categorical factors, and much more.

• Directly optimize for Bayesian a-priori variance ratios in designs with hard-to-change factors.

• High-performance model selection using SAMS (simulated annealing model selection) (Wolters and Bingham, 2012).

Getting started

Install this package using pip

pip install pyoptex

Documentation

The documentation for this package can be found at here

Create your first design

See the documentation on Your first design

Analyze your first dataset

See the documentation on Your first dataset

License

BSD-3 clause, meaning you can use and alter it for any purpose, open-source or commercial! However, any open-source contributions to this project are much appreciated by the community.

Contributing

Any ideas, bugs and features requests can be added as an issue. Any direct code contributions can be added via pull requests.