process-improve#

Designed experiments, multivariate models and process monitoring in Python: the companion package to the free textbook Process Improvement using Data.

Four designs from one generate_design call: a definitive screening design, a central composite design, a constrained I-optimal design and a MaxPro space-filling design. Four designs from one generate_design call: a definitive screening design, a central composite design, a constrained I-optimal design and a MaxPro space-filling design. 
pip install process-improve

Get started Try it in your browser User guide API reference The textbook

What it does#

Designed experiments

Factorial, response surface, definitive screening, OMARS, optimal and mixture designs from one generate_design call, and the analysis of the results.

Plan an experimental programme

Judge a design before you run it

Efficiency, prediction variance, correlation between coefficients and bias from omitted terms, all computed for the model you intend to fit.

Evaluate a design

Multivariate models

PCA, PLS and TPLS in scikit-learn style, with scores, loadings, contributions, and Hotelling’s T² and SPE limits for spotting unusual observations.

Latent variable models

Process monitoring

Shewhart, EWMA and CUSUM control charts, and adaptive PCA and PLS models that update as new samples arrive.

Monitoring reference

Batch processes

A batch bioreactor simulator, batch trajectory features, and mid-course correction of a batch that is still running.

The batch simulator

Sensory panels

Validate descriptive panel data, check each assessor, and relate the attribute scores back to the products.

Descriptive panel data

Try it without installing anything#

The designed-experiments app in a browser: choosing factors, generating the design, and reading the analysis of the results.

Factorial, the designed-experiments app, runs process-improve inside your browser. Design an experiment, download the run sheet, fill in your results, and upload it for the analysis, with nothing to install.

Open the app, or browse the design catalog: the designs of textbooks, handbooks and papers, and those the library makes, each opened in the app with a click.

Sixteen designs, one function call#

Sixteen design families, each drawn as its runs: supersaturated, Plackett-Burman, fractional factorial, definitive screening, full factorial, central composite, Box-Behnken, OMARS, Taguchi, D-optimal, constrained I-optimal, constrained mixture, Latin hypercube, uniform, MaxPro and Sobol designs. Sixteen design families, each drawn as its runs: supersaturated, Plackett-Burman, fractional factorial, definitive screening, full factorial, central composite, Box-Behnken, OMARS, Taguchi, D-optimal, constrained I-optimal, constrained mixture, Latin hypercube, uniform, MaxPro and Sobol designs.

Screening, response surface, optimal, mixture and space-filling designs all come from generate_design(factors, design_type=...). Each construction is checked against its published definition by the test suite, so a change that breaks a design fails the tests.

See every design family

Learn the methods#

Process Improvement using Data, a free online textbook by Kevin G. Dunn, beside one of its Python examples with the Run in browser button and the plot a click draws.

The free textbook Process Improvement using Data, by the same author, explains the statistics behind the package: designed experiments, process monitoring, regression and latent variable models. Its Python examples use process-improve, and over 180 of them run in your browser with a click.

Read the textbook