Advanced Features: Async Evaluation and CLI

This section covers Cadence's advanced capabilities, including asynchronous evaluation of programs and the built-in command-line interface.

1. Asynchronous Evaluation

Cadence can leverage Python's asyncio to run multiple evaluations concurrently, reducing overall runtime for batch assessments.

import asyncio
from src.evaluator import Evaluator
from src.tasks.tsp_task import TSPTask

async def async_eval():
    evaluator = Evaluator()
    task = TSPTask(n_cities=8)
    seeds = list(range(5))
    # Run evaluations in parallel threads
    results = await asyncio.gather(*[
        asyncio.to_thread(evaluator.evaluate_code, task.baseline_program, task, [s])
        for s in seeds
    ])
    for i, res in enumerate(results):
        print(f"Seed {i}: cost={res.cost:.2f}, feasible={res.feasible}")

asyncio.run(async_eval())

2. Command-Line Interface (CLI)

Cadence provides a simple CLI via main.py for running experiments without modifying code.

# Run a TSP evolution for 10 generations and save results
gitbash
python main.py --task tsp --generations 10 --output evolution_results.sqlite

CLI Options

• --task: Task name (e.g., tsp)
• --generations: Number of evolution generations
• --population: Population size per generation (default: 5)
• --output: Path to SQLite output database
• --seeds: Comma-separated random seeds for evaluation