Optienv: Introduction
Optienv is a lightweight, simulator‑friendly framework for multi‑objective optimization (MOO). It wraps external models with a simple CSV interface and runs evolutionary algorithms—currently NSGA‑II and NSGA‑III—at scale.
Design goals
Reproducible experiments Single-file per‑seed histories, optional checkpoints/resume, and fixed seeds.
Simulator integration CSV adapter reads decision vectors and writes objectives—no refactoring in your model.
Many‑objective capability NSGA‑III with reference directions for high‑dimensional objectives (≥ 4).
Actionable analysis Global Pareto front extraction and normalized hypervolume (HV) computation, including density control via
--epsilon.
Key features
Two algorithms - NSGA‑II (dominance ranking + crowding distance) - NSGA‑III (dominance ranking + reference‑direction niching)
CSV model adapter - Writes
variable_values.csvfor your wrapper - Expectsobjective_values.csvback from the wrapper - Converts maximize objectives to minimization internallyResults management - One history CSV per seed:
results/history_seed{SEED}.csv- Optional final CSVs: population, fitness, non‑dominated (nd) maskAnalysis - front: global non‑dominated set (optionally thinned by ε‑boxes) - hypervolume: normalized HV per generation (wide format)