Skip to content

INTEGRATIONS.md

Latest commit

 

History

History
45 lines (31 loc) · 1.31 KB

INTEGRATIONS.md

File metadata and controls

45 lines (31 loc) · 1.31 KB
Raw

optimization Integration Guide

Install from the repository:

pip install "git+https://github.com/FlossWare/optimization.git"

Python usage

The package exposes a genetic algorithm and an explicit @optimize decorator. The caller supplies a fitness function; optimization does not invoke models, classify tasks, or orchestrate workflows.

from optimization import GeneticOptimizer, Individual, ParamBounds, optimize

bounds = [
    ParamBounds("temperature", 0.0, 2.0),
    ParamBounds("top_p", 0.0, 1.0),
]

optimizer = GeneticOptimizer(params=bounds, seed=7)

def fitness(candidate: Individual) -> float:
    # Replace with a deterministic, caller-owned objective.
    return -(candidate.genes["temperature"] - 0.5) ** 2

best = optimizer.evolve(fitness, generations=20)
print(best.genes)

@optimize(param_bounds={"temperature": (0.0, 2.0)}, seed=7)
def score(temperature: float) -> float:
    return -(temperature - 0.5) ** 2

result = score()
print(result.best_params, result.fitness)

Boundary

  • optimization owns reusable optimization algorithms.
  • strategy owns decision-selection algorithms.
  • evaluation assesses outcomes and evidence.
  • Loom owns execution semantics and orchestration.

Optimization must not override policy, authorization, safety, feasibility, or budget constraints.