Getting started¶

Installation¶

python -m pip install industrialstats

The supported Python range is 3.11 through 3.14. Core dependencies are NumPy, pandas, SciPy, statsmodels, scikit-learn, Matplotlib, seaborn, Plotly, openpyxl, and DataExcept.

Defining factors¶

Every design is built from Factor objects. A factor has a name, its levels, and a type of either "continuous" or "categorical".

from industrialstats.designs.base import Factor

temperature = Factor("temperature", [180, 220], factor_type="continuous")
material = Factor("material", ["ABS", "PP"], factor_type="categorical")

Generating a full factorial design¶

from industrialstats.designs.factorial import FactorialDesign

design = FactorialDesign(
    factors=[temperature, material],
    replicates=2,
    randomize=True,
    seed=42,
)
matrix = design.generate_design()
print(matrix)

Always pass a seed

Randomization is seedable throughout the package. Passing seed makes a run reproducible, which matters both for auditing an experiment and for writing deterministic tests.

Fractional factorials and aliasing¶

When the full factorial is too large, use a regular fraction and inspect what it costs you:

from industrialstats.designs.fractional_factorial import FractionalFactorialDesign

factors = [Factor(name, [-1, 1]) for name in "ABCDEFG"]
design = FractionalFactorialDesign(factors, fraction="1/8", randomize=False)

design.generate_design()
print(design.resolution_analysis())
print(design.alias_structure()["A"])

The alias structure tells you which effects are indistinguishable in the fraction you chose. Read it before running the experiment, not after.

Response surface methodology¶

from industrialstats.designs.response_surface import ResponseSurfaceDesign

factors = [
    Factor("temperature", [180, 220], factor_type="continuous"),
    Factor("pressure", [10, 20], factor_type="continuous"),
]

design = ResponseSurfaceDesign(factors, design_type="CCD", center_points=4)
print(design.generate_design())

Analysing results¶

Once responses are collected, fit a model and produce an ANOVA table:

from industrialstats.analysis.anova import ANOVAAnalysis

analysis = ANOVAAnalysis(data, "Response")
analysis.fit_model("Response ~ temperature + pressure")
print(analysis.anova_table_calculation(typ=2))

Error handling¶

Data-loading and export boundaries raise DataExcept exception types rather than bare OSError or ValueError, and preserve the original exception as the cause:

from dataexcept import FileWriteError
from industrialstats.utils.export import export_to_csv

try:
    export_to_csv(matrix, "missing-directory/design.csv")
except FileWriteError as exc:
    print(exc.path, exc.original)

Mathematical precondition failures — an invalid factor level, a singular design matrix — remain explicit ValueError and related exceptions.