run.py

# run.py
import numpy as np
import pandas as pd
import time

from rosenbrock import rosenbrock
from sd import sd
from bfgs import bfgs
from lbfgs import lbfgs

# ============================================
ALGORITHM = 'SD'  # 'SD', 'BFGS', 'L_BFGS'
DIM = 200  # 维度
N_POINTS = 100  # 测试点数
m = 10  # L-BFGS 用的 m
success_t=1e-8#收敛的判断标准
# =====================================================

np.random.seed(0)
floatX = np.longdouble#至关重要

# 选择优化器
if ALGORITHM == 'SD':
    opt = sd
elif ALGORITHM == 'BFGS':
    opt = bfgs
elif ALGORITHM == 'L_BFGS':
    opt = lambda f, x0: lbfgs(f, x0, m=m)

# 生成随机初始点
range_val = 3
X0 = 1 + 2 * range_val * (np.random.rand(N_POINTS, DIM) - 0.5)
X0 = np.asarray(X0, dtype=floatX)

records = []

print(f"开始测试 {ALGORITHM} | 维度 {DIM} | 测试点 {N_POINTS} 个")
print("-" * 70)

for i in range(N_POINTS):
    x0 = X0[i]
    t0 = time.time()
    x, fval, it = opt(rosenbrock, x0)[:3]
    t = time.time() - t0

    dist = np.linalg.norm(x - np.ones(DIM, dtype=floatX))
    err = abs(fval)
    success = dist < success_t and err < success_t

    records.append([i + 1, DIM, ALGORITHM, it, round(t, 4), dist, err, success])

    status = "成功" if success else "失败"
    print(f"点 {i + 1:3d} → {it:5d} 次迭代  {t:6.3f} 秒  误差 {err:.2e}  {status}")

# 保存结果
df = pd.DataFrame(records, columns=[
    'id', 'dim', 'algo', 'iter', 'time_s', 'dist', 'error', 'success'])

filename = f"result\{ALGORITHM}_dim{DIM}_{N_POINTS}pts.csv"
df.to_csv(filename, index=False, float_format='%.16e')

success_rate = df.success.mean()
print("-" * 70)
print(f"全部完成！")
print(f"成功率：{success_rate:.1%}  ({df.success.sum()}/{N_POINTS})")
print(f"平均迭代次数：{df.iter.mean():.1f}")
print(f"平均时间：{df.time_s.mean():.3f} 秒")
print(f"结果已保存 → {filename}")