standard tier · 72 briefs · 1,368 runs · results db2e24e

CityFormGen

Generative and optimisation methods for middle housing, compared under matched compute

RESEARCH GENERATIVE-DESIGN BENCHMARK - NOT CONSTRUCTION DOCUMENTATION OR CODE CERTIFICATION.

Pareto explorer

Six objectives, no single score. Pick a brief and any two objectives: every feasible design every method found appears, and the designs nondominated on those two axes are ringed.

Loading designs…

Who builds the pooled front

Per brief, all methods' feasible designs are pooled and the 6-D nondominated set kept. Share of that front contributed by each method, mean over briefs, with 95% bootstrap interval.

Heuristic-seeded NSGA-II26% ±
NSGA-II24% ±
Random sketch + repair22% ±
Heuristic13% ±
CP-SAT8% ±
Simulated annealing7% ±
Random search0% ±
Share of the pooled 6-D front, by method.

Coverage between fronts

C(A, B): the share of B's feasible designs weakly dominated by some design of A. Not symmetric. The twelve largest pairs are shown.

A coversBC(A,B)briefs
NSGA-IIRandom search0.8714
Random sketch + repairRandom search0.7414
Simulated annealingRandom search0.6014
CP-SATHeuristic-seeded NSGA-II0.4727
CP-SATRandom search0.4514
NSGA-IISimulated annealing0.3039
Random sketch + repairSimulated annealing0.2940
NSGA-IIRandom sketch + repair0.2843
NSGA-IIHeuristic-seeded NSGA-II0.2436
Random sketch + repairHeuristic-seeded NSGA-II0.2437
Random sketch + repairNSGA-II0.2243
CP-SATSimulated annealing0.2129