FORMULA-X - Frequently Asked Questions
Those tools fit a quadratic surface and apply Derringer-Suich
desirability. FORMULA-X adds Bayesian optimization, probabilistic
design space (Pr(spec met) maps), Pareto fronts, robust optima,
constraint-aware search, and an honest RSM-vs-ML cross-validated
comparison. The classical workflow is still here, it is just one
of several tools rather than the only one.
Full and fractional factorial, Plackett-Burman, Box-Behnken,
central composite (face-centered, rotatable, inscribed),
D-optimal, simplex mixture, and Latin hypercube. Or upload your
own design CSV.
Quadratic RSM needs at least
1 + 2k + k(k-1)/2 + 1 rows
for k continuous factors. GP and GBM can run with fewer rows but
will report wide uncertainty - which FORMULA-X surfaces honestly
via the probabilistic design-space map rather than hiding behind
a single point estimate.
Yes. Constraints are expressed in plain symbolic form using factor
and response names, and are enforced inside Pareto search,
desirability, and Bayesian optimization. Mixture sum constraints
are handled the same way (e.g.
w_lecithin + w_GMS = 1).
A specific row of factor values together with the acquisition
score that earned it (Expected Improvement by default). Run the
experiment in the lab, upload the observed response back into
FORMULA-X, and the surrogate updates. After a few rounds you can
compare the BO trajectory to the classical RSM optimum - this is
the methodological framing that turns a routine DoE study into a
genuine methods contribution.
From the Export tab: a publication-ready PDF report (methods,
metrics, Pareto figure, design-space heatmap, BO trajectory,
references), a zipped CSV bundle of every artifact, and per-figure
PNG and SVG downloads. PDF/CSV exports arrive in milestone M6.