baseline

2026-09-12 16:58:25 +08:00
parent 1fa2264a17
commit 892451dca9
36 changed files with 8767 additions and 0 deletions
@@ -0,0 +1,148 @@
+"""统一入口：一次跑完 1-4 问的"最优 vs 基线"对比并生成 results/。
+
+用法（仓库根目录）::
+
+    uv run python -m project.run                  # 四问全部
+    uv run python -m project.run --question 3     # 只跑第三问
+    uv run python -m project.run --quick          # 缩小算例/基准集规模
+"""
+
+from __future__ import annotations
+
+import argparse
+import time
+
+from project.common import RESULTS_DIR, markdown_table, write_text
+from project.q1 import compare as q1_compare
+from project.q2 import compare as q2_compare
+from project.q3 import compare as q3_compare
+from project.q4 import compare as q4_compare
+
+RUNNERS = {
+    1: q1_compare.run,
+    2: q2_compare.run,
+    3: q3_compare.run,
+    4: q4_compare.run,
+}
+
+STRUCTURE = """\
+project/
+├── common.py            公共层：方法描述/指标聚合/结果落盘/闭环装配
+├── q1/                  问题一：直径与直径圆覆盖
+│   ├── optimal.py       生产实现：半平面交 + 顶点对直径 + MEC
+│   ├── baseline.py      基线：网格采样 + bbox 对角线代理
+│   ├── reference.py     参考：大框半平面裁剪 + 精确直径/MEC
+│   ├── cases.py         固定示例（examples/q1）+ 随机交会算例
+│   └── compare.py       同案配对对比
+├── q2/                  问题二：第二检测点选择
+│   ├── optimal.py       生产实现：保证接收盘内最小最坏后验半径
+│   ├── baseline.py      基线：垂直 1000 m / MEC 圆心
+│   ├── reference.py     参考：保证接收盘内网格枚举
+│   ├── region.py        K1 统一构造
+│   ├── cases.py         随机首次示向度算例
+│   └── compare.py
+├── q3/                  问题三：全向闭环
+│   ├── optimal.py       bayes（生产默认策略）
+│   ├── baseline.py      grid（49 点网格）/ seven（七点批量）
+│   └── compare.py
+├── q4/                  问题四：混合源闭环
+│   ├── optimal.py       directional（定向扩展）
+│   ├── baseline.py      bayes / grid 直接搬用
+│   └── compare.py
+├── run.py               统一入口（本文件）
+└── results/             自动生成：q1.md/json … q4.md/json + summary.md
+"""
+
+
+def _q1_headline(payload: dict) -> list[str]:
+    agg = payload["aggregate_random"]
+    return ["1", "直径平均误差 / 覆盖准确率",
+            f"{agg['diameter_mae_optimal_m']:.2f} m / "
+            f"{agg['coverage_accuracy_optimal']*100:.0f}%",
+            f"{agg['diameter_mae_baseline_m']:.2f} m / "
+            f"{agg['coverage_accuracy_baseline']*100:.0f}%"]
+
+
+def _q2_headline(payload: dict) -> list[str]:
+    agg = payload["aggregate"]
+    opt = agg["optimal"]
+    base = agg["baseline_mec"]
+    return ["2", "保证接收率 / 最坏后验均值",
+            f"{opt['guaranteed_rate']*100:.0f}% / "
+            f"{opt['worst_posterior_mean_m']:.1f} m",
+            f"{base['guaranteed_rate']*100:.0f}% / "
+            f"{base['worst_posterior_mean_m']:.1f} m"]
+
+
+def _q3_headline(payload: dict) -> list[str]:
+    agg = payload["aggregate"]["variants"]
+    opt, base = agg["optimal"], agg["baseline"]
+    paired = payload["aggregate"]["paired"].get("optimal_vs_baseline", {})
+    return ["3", "全清率 / T 均值（相对基线）",
+            f"{opt['cleared_ratio_mean']*100:.0f}% / "
+            f"{opt['virtual_time_mean_s']:.0f} s "
+            f"({paired.get('mean_relative_change', 0)*100:+.0f}%)",
+            f"{base['cleared_ratio_mean']*100:.0f}% / "
+            f"{base['virtual_time_mean_s']:.0f} s"]
+
+
+def _q4_headline(payload: dict) -> list[str]:
+    agg = payload["aggregate"]["variants"]
+    opt, base = agg["optimal"], agg["baseline"]
+    return ["4", "全清率 / 缺席误判（问题三策略搬用）",
+            f"{opt['cleared_ratio_mean']*100:.0f}% / "
+            f"{opt['absent_misjudged_total']}",
+            f"{base['cleared_ratio_mean']*100:.0f}% / "
+            f"{base['absent_misjudged_total']}"]
+
+
+def _write_summary(payloads: dict[int, dict]) -> None:
+    lines = ["# 1-4 问最优 vs 基线：对比总览", "",
+             "本节所有数字由 `uv run python -m project.run` 在同机同口径下生成，",
+             "逐问明细见 `q1.md` … `q4.md`。", ""]
+    rows = []
+    if 1 in payloads:
+        rows.append(_q1_headline(payloads[1]))
+    if 2 in payloads:
+        rows.append(_q2_headline(payloads[2]))
+    if 3 in payloads:
+        rows.append(_q3_headline(payloads[3]))
+    if 4 in payloads:
+        rows.append(_q4_headline(payloads[4]))
+    lines.append(markdown_table(["问", "核心指标", "最优实现", "基线实现"], rows))
+    lines += ["", "## 方法映射", "",
+              "| 问 | 最优实现 | 基线实现 | 独立参考 |",
+              "| --- | --- | --- | --- |",
+              "| 1 | 半平面交 + 顶点对直径 + MEC | 网格采样 + bbox 对角线代理 | 大框半平面裁剪 + 精确直径/MEC |",
+              "| 2 | 保证接收盘内最小最坏后验半径 | 垂直 1000 m / MEC 圆心 | 保证接收盘内网格枚举 |",
+              "| 3 | bayes（粒子滤波 + 信息增益 + 滚动调度） | grid（49 点网格）/ seven（七点批量） | 冻结基准集 + 逐微秒计时复算 |",
+              "| 4 | directional（联合信念 + 后备清除） | bayes / grid 直接搬用 | 冻结基准集 + 真值全清校验 |",
+              "", "## 目录结构", "", "```text", STRUCTURE.rstrip(), "```", ""]
+    write_text(RESULTS_DIR / "summary.md", "\n".join(lines))
+
+
+def main(argv=None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument("--question", choices=("1", "2", "3", "4", "all"),
+                        default="all")
+    parser.add_argument("--quick", action="store_true",
+                        help="缩小随机算例数/冻结基准集规模")
+    args = parser.parse_args(argv)
+
+    selected = [1, 2, 3, 4] if args.question == "all" else [int(args.question)]
+    payloads: dict[int, dict] = {}
+    for question in selected:
+        start = time.perf_counter()
+        print(f"[Q{question}] 运行对比 …", flush=True)
+        payloads[question] = RUNNERS[question](quick=args.quick)
+        print(f"[Q{question}] 完成，用时 {time.perf_counter() - start:.1f}s",
+              flush=True)
+    if selected == [1, 2, 3, 4]:
+        _write_summary(payloads)
+        print(f"总览已写入 {RESULTS_DIR / 'summary.md'}")
+    print(f"结果目录：{RESULTS_DIR}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())