Release SONAPSE-ORIGIN v0.2.0: Neural Organisation Discovery, Causal Emergence, PCI_LZ, and CTC
5b4a3bd verified | import numpy as np | |
| import pytest | |
| from sonapse_origin.evolution.map_elites import MAPElitesArchive | |
| from sonapse_origin.evolution.nsga2 import NSGAParetoOptimizer | |
| from sonapse_origin.evolution.novelty_search import NoveltySearchArchive | |
| from sonapse_origin.evolution.search_engine import ArchitectureDiscoveryEngine | |
| from sonapse_origin.structure.genome import TopologyFamily | |
| from sonapse_origin.structure.topology import TopologyGenerator | |
| def test_map_elites_archive(): | |
| archive = MAPElitesArchive(dim_x=5, dim_y=5) | |
| gen = TopologyGenerator(seed=1) | |
| g1 = gen.generate(TopologyFamily.FEEDFORWARD, num_nodes=8) | |
| g2 = gen.generate(TopologyFamily.RECURRENT, num_nodes=8) | |
| added1 = archive.add_candidate(g1, fitness=1.5, behavior_descriptor=(0.2, 0.4)) | |
| assert added1 | |
| assert archive.coverage() > 0.0 | |
| # Better candidate in same cell replaces | |
| added2 = archive.add_candidate(g2, fitness=2.0, behavior_descriptor=(0.22, 0.42)) | |
| assert added2 | |
| assert archive.max_fitness() == 2.0 | |
| def test_nsga2_pareto(): | |
| nsga = NSGAParetoOptimizer() | |
| objs = [ | |
| np.array([1.0, 2.0]), | |
| np.array([2.0, 1.0]), | |
| np.array([0.5, 0.5]), # Dominated by both | |
| ] | |
| fronts = nsga.fast_non_dominated_sort(objs) | |
| assert len(fronts) >= 2 | |
| assert set(fronts[0]) == {0, 1} | |
| assert fronts[1] == [2] | |
| def test_novelty_archive(): | |
| nov = NoveltySearchArchive(k_neighbors=2, novelty_threshold=0.3) | |
| p1 = np.array([0.0, 0.0]) | |
| p2 = np.array([1.0, 1.0]) | |
| nov.maybe_add_to_archive(p1, novelty=1.0) | |
| nov.maybe_add_to_archive(p2, novelty=1.0) | |
| # Point close to p1 has low novelty | |
| n_close = nov.compute_novelty(np.array([0.05, 0.05])) | |
| # Point far away has high novelty | |
| n_far = nov.compute_novelty(np.array([5.0, 5.0])) | |
| assert n_far > n_close | |
| def test_architecture_discovery_engine(): | |
| engine = ArchitectureDiscoveryEngine(population_size=6, seed=42) | |
| def dummy_eval(genome): | |
| mat = genome.to_matrix() | |
| density = float(np.mean(mat != 0)) | |
| mean_w = float(np.mean(np.abs(mat))) | |
| fitness = density * 2.0 + mean_w | |
| b_desc = (float(np.clip(density, 0.0, 1.0)), float(np.clip(mean_w, 0.0, 1.0))) | |
| objs = np.array([fitness, density, mean_w, 1.0, 1.0, -1.0, -1.0]) | |
| return fitness, b_desc, objs | |
| res = engine.evolve_step(dummy_eval) | |
| assert res["max_fitness"] > 0.0 | |
| assert res["map_elites_coverage"] > 0.0 | |