mirror of
https://github.com/blackboxprogramming/lucidia.git
synced 2026-03-16 23:57:09 -05:00
Alexa Amundson
855585cb0e
Synced from BlackRoad-OS-Inc/blackroad-operator/orgs/personal/lucidia BlackRoad OS — Pave Tomorrow. RoadChain-SHA2048: fe729062952871e7 RoadChain-Identity: alexa@sovereign RoadChain-Full: fe729062952871e77147cf6d938b799096e87d9024d7005a14c9e209e12e8ad0c825b624c7bc649fc7eeb4c284fdcab8231af77980065cc04d9f36fca479ffc2346ed3c1b73de6f240d8f9485f47c995ad5b81142f7179b84932c67914dff1c08db039349ba28fca36cb57688093bf0199268dd1c2f3448c9383000bc77cc9663066ff57b834370afc8838b18466ea9029908018b961555cccaabf2ce21649cf3cabc7f64bdcc4abdf2da259b210c342835a2cecf92bdd3b4e109b4d6e622f6934e13b2b123607bd61ce3d0f20454c9ab594f9284cffe18716619c52db57ce5f4ee2856cb96e1fa3748fe1fe65435bec297c5ab3ab58d570ec1064aea29931dd
60 lines
2.6 KiB
Python
60 lines
2.6 KiB
Python
"""
|
|
Capability Optimizer for Mirror Engine.
|
|
|
|
This script performs a simple random search over mirror engine parameters to maximise capability
|
|
defined as the harmonic mean of reach and stability. It leverages the run_mirror_engine function
|
|
from mirror_engine.py and summarises results.
|
|
"""
|
|
import numpy as np
|
|
import random
|
|
from mirror_engine import run_mirror_engine
|
|
|
|
def evaluate_params(params):
|
|
history = run_mirror_engine(iterations=params.get('iterations', 20),
|
|
target=params.get('target', 0.5),
|
|
threshold=params.get('threshold', 0.1),
|
|
step_init=params.get('step_init', 1.0),
|
|
min_step=params.get('min_step', 0.01),
|
|
max_step=params.get('max_step', 10.0))
|
|
# compute reach: fraction of aggregated values within reach_threshold of target
|
|
aggregates = np.array([rec['aggregate'] for rec in history], dtype=float)
|
|
step_sizes = np.array([rec['step_size'] for rec in history], dtype=float)
|
|
target = params.get('target', 0.5)
|
|
reach_threshold = params.get('reach_threshold', 0.1)
|
|
reach = float(np.mean(np.abs(aggregates - target) <= reach_threshold))
|
|
# compute stability: inverse of normalised step variance (lower variance implies stability)
|
|
step_std = float(np.std(step_sizes))
|
|
stability = 1.0 / (1.0 + step_std)
|
|
capability = 0.0
|
|
if (reach + stability) > 0:
|
|
capability = 2.0 * reach * stability / (reach + stability)
|
|
return {'reach': reach, 'stability': stability, 'capability': capability, 'params': params}
|
|
|
|
def random_search(num_samples=10):
|
|
"""Perform random search over parameter space to find configurations with high capability."""
|
|
results = []
|
|
for _ in range(int(num_samples)):
|
|
params = {
|
|
'iterations': random.randint(10, 30),
|
|
'target': random.uniform(0.1, 0.9),
|
|
'threshold': random.uniform(0.05, 0.2),
|
|
'step_init': random.uniform(0.1, 5.0),
|
|
'min_step': 0.01,
|
|
'max_step': 10.0,
|
|
'reach_threshold': random.uniform(0.05, 0.2)
|
|
}
|
|
res = evaluate_params(params)
|
|
results.append(res)
|
|
results_sorted = sorted(results, key=lambda x: x['capability'], reverse=True)
|
|
return results_sorted
|
|
|
|
if __name__ == "__main__":
|
|
search_results = random_search(20)
|
|
best = search_results[0] if search_results else None
|
|
if best:
|
|
print(f"Best capability: {best['capability']:.3f}")
|
|
print(f"Parameters: {best['params']}")
|
|
print(f"Reach: {best['reach']:.3f}, Stability: {best['stability']:.3f}")
|
|
else:
|
|
print("No results")
|