Skip to content

tradespace

For an analyst with a table of evaluated designs. The paper makes the argument for the method; the project page links it, these docs and the package.

tradespace learns the shape of a design trade space from evaluated designs. Give it a table of designs, each scored on several metrics where larger is better, and it fits a frontier model in the posynomial invariant form: a sum of terms, where a term (a product of powers of the metrics) is one trade-off. It then tells you which metrics contend, with what elasticity, where each design sits relative to the fitted reference surface, and where the table has white space.

A first fit, on a synthetic table of 300 designs scored on 3 metrics:

from tradespace import (
    HyperbolicTradeSpace,
    make_frontier_data,
    trade_off_report,
)
from tradespace.plotting import plot_log_plane

X, info = make_frontier_data(300, 3, random_state=0)
ts = HyperbolicTradeSpace(temperature=20.0).fit(X)
report = trade_off_report(ts, X)
print(report.lemma_name, report.contending_set)
print(report.exponents.round(2))
X2 = X[:, :2]
ts2 = HyperbolicTradeSpace(temperature=20.0).fit(X2)
plot_log_plane(ts2, X2, band=0.5)

Designs in the log plane of two metrics: the frontier is a straight line, and the designs scatter below it, a few just beyond it, and the nearest ones lie in a grey band under the line. Designs in the log plane of two metrics: the frontier is a straight line, and the designs scatter below it, a few just beyond it, and the nearest ones lie in a grey band under the line.

The same fit on two of the metrics: the line is the fitted frontier, the designs fall below it or just beyond it, and the grey band holds those nearest to it; Getting started walks through the output.

It is a scikit-learn estimator (fit, transform, inverse_transform, score_samples) in unsupervised learning, beside PCA: PCA finds where the data spread, this finds where they stop, and the two chain.