Examples gallery

Every entry below is a real, runnable script under examples/python/ — not a snippet reproduced for the docs. Run any of them directly:

py-sezgi/.venv/bin/python examples/python/oop/gwo.py

Prefer an interactive, cell-by-cell walkthrough instead? See the Notebooks section for four executed Jupyter notebooks covering the same class-first surface.

This gallery is a curated subset (12 of the full catalog's ~51 scripts): six representative picks from the 19 parameterized OOP algorithm twins, the five engine-hosted authoring examples (examples/python/oop/engine/), and the feature-selection data recipe. See examples/README.md for the full 17-algorithm catalog (pure-script / OOP-twin / spec-file triplets) this gallery draws from.

Algorithm twins (sezgi.AskTellAlgorithm, ask/tell style)

Each of these ports a pure-Python metaheuristic script onto sezgi.AskTellAlgorithm — same update equations, same RNG draw order, bit-for-bit matched against its pure-script twin (test_examples_oop_parity.py's 17-pair gate). All four below solve sezgi.bbob(1, dim, 1) (Sphere) at budget 2000, seed 42.

Example	File	Category	What it teaches
Grey Wolf Optimizer	examples/python/oop/gwo.py	Swarm intelligence	Leader-following update with a literal a = 2 - 2*progress explore-to-exploit decay schedule (see Exploration vs exploitation)
Cuckoo Search	examples/python/oop/cs.py	Swarm intelligence (Lévy flight)	Lévy-flight step generation via a closed-form Mantegna (1994) algorithm, stdlib-only (math.gamma in place of the Rust component's Lanczos gamma)
Harris Hawks Optimization	examples/python/oop/hho.py	Swarm intelligence	A multi-phase (exploration/exploitation) update rule with a generational replacer
Teaching-Learning-Based Optimization	examples/python/oop/tlbo.py	Non-swarm metaheuristic	A population update driven by a "teacher" and pairwise "learner" interactions — no leader/pheromone/velocity metaphor at all

Engine-hosted authoring (sezgi.Algorithm family, class-first)

These seven (five under examples/python/oop/engine/, plus two more from examples/python/oop/ demonstrating the same engine-hosted surface) each demonstrate ONE distinct way to author against the engine-hosted class surface (sezgi.Algorithm/PopulationAlgorithm/LocalSearch/ Problem) — Tutorials 3, 4, and 6 walk through the same hooks these scripts exercise.

Example	File	Category	What it teaches
DE/rand/1, vary() only	examples/python/oop/custom_de_variant.py	Population algorithm	The smallest possible PopulationAlgorithm subclass: one method, inherits the base's tournament select()
DE/rand/2/bin, two donor vectors	examples/python/oop/engine/custom_de.py	Population algorithm	A second, more elaborate vary() — two difference terms plus binomial crossover — read alongside custom_de_variant.py for the mutation/crossover split's two ends
Simplified PSO, full generate()	examples/python/oop/engine/custom_pso_variant.py	Full Algorithm override	No select()/vary() split to lean on, plus per-instance state (velocities, personal/global bests) carried across generate() calls on self
Random 2-opt local search	examples/python/oop/engine/local_search_2opt.py	Local search	A LocalSearch subclass (neighbor() only) over a hand-authored PERMUTATION-typed Problem — a small in-file TSP instance, no vendored data file
Custom perturbation local search	examples/python/oop/custom_local_search.py	Local search	The Float-space counterpart to local_search_2opt.py — perturbs real coordinates instead of reversing a tour segment
Rastrigin problem authoring	examples/python/oop/engine/custom_problem_rastrigin.py	Problem authoring	Defines the PROBLEM side only (pure math, no numpy) and solves it with a stock sezgi.GreyWolfOptimizer — the other half of the class-first surface from every row above
Mixed-space Problem + Algorithm	examples/python/oop/engine/mixed_space_tuning.py	Mixed-space authoring	A Space(Float, Categorical, Int) problem paired with an Algorithm that varies all three sub-blocks and vetoes any other space via validate_space() — see Tutorial 6

Data recipes

Example	File	Category	What it teaches
Feature selection	examples/python/oop/feature_selection.py	Data recipe	sezgi.recipes.FeatureSelection recovering a known 3-of-8 informative-column mask from a synthetic dataset via GeneticAlgorithm's Binary auto-dispatch — see Tutorial 5

Running these yourself

Every script above is stdlib/numpy-only (no sklearn/scipy anywhere in this project's Python surface), deterministic under a fixed seed, and ends by printing an evals_used=... best_f=... gap=... line — the same metrics-line convention this whole site's tutorials use. Each is gated by its own anchored pytest (py-sezgi/tests/test_examples_oop_parity.py, test_oop_families.py, test_examples_engine.py, or test_feature_selection_example.py), so a script's printed numbers are never allowed to silently drift from what is committed here.