Command-line interface¶

Installing the package provides an industrialstats command for generating designs and running selected analyses without writing a script.

industrialstats --help

Every subcommand accepts -o/--output to write CSV instead of printing to standard output.

Factor specification¶

The factorial, fractional and screening subcommands take repeated -f/--factor arguments in the form NAME=level1,level2. The crd and rcbd subcommands instead take repeated -t/--treatment arguments, one per treatment level, because they describe a single treatment factor:

industrialstats factorial -f "Temperature=180,220" -f "Pressure=10,20"

Generating designs¶

Full factorial¶

industrialstats factorial \
  -f "Temperature=180,220" \
  -f "Pressure=10,20" \
  --replicates 2 \
  --center-points 3 \
  -o design.csv

Fractional factorial¶

Specify either a --fraction or explicit --generator strings:

industrialstats fractional -f "A=-1,1" -f "B=-1,1" -f "C=-1,1" --fraction 1/2

Completely randomized design¶

industrialstats crd -t A -t B -t C --replicates 4 --seed 42

Randomized complete block design¶

industrialstats rcbd \
  -t A -t B -t C \
  -b Day1 -b Day2 -b Day3 \
  --blocking-factor Day \
  --seed 42

Screening¶

--design pb builds a Plackett-Burman design; --design dsd builds a definitive screening design.

industrialstats screening -f "A=-1,1" -f "B=-1,1" -f "C=-1,1" --design pb --seed 42

Warning

The dsd construction is experimental and scheduled for statistical correction. Prefer pb for production screening.

Analysis¶

ANOVA¶

industrialstats anova \
  --data results.csv \
  --response Yield \
  --formula "Yield ~ Temperature * Pressure" \
  --typ 2

--typ selects the sum-of-squares type (1, 2, or 3). Type II is the usual choice for balanced designs without significant interactions; type III is used when interactions are present.

Power analysis¶

industrialstats power --analysis t-test --effect-size 0.5 --power 0.8

Supply exactly two of --effect-size, --power, and --sample-size; the command solves for the third.

For a one-way ANOVA:

industrialstats power --analysis anova --effect-size 0.4 --power 0.8 --n-groups 4

Stepwise model fitting¶

printf 'y,A,B\n1,0,0\n2,0,1\n3,1,0\n4,1,1\n' > model.csv
industrialstats model \
  --data model.csv \
  --response y \
  --entry-threshold 0.01 \
  --removal-threshold 0.2

Stepwise selection inflates significance

p-values from a stepwise search are optimistic because the model was chosen using the same data. Treat the selected terms as a hypothesis to confirm, not as a validated model.

Reproducibility¶

The crd, rcbd and screening subcommands accept --seed. Record it with your results: it is what allows the exact run order to be reconstructed later.

factorial and fractional randomize without a seed

FactorialDesign randomizes by default, but the factorial and fractional subcommands expose no --seed flag, so their run order cannot currently be reproduced from the command line. Save the generated CSV with -o, or use the Python API, which accepts seed.

Errors¶

File reads and writes raise structured DataExcept errors that name the offending path, and the command exits with a non-zero status.