E.0 选修模块实操工作坊

选修模块实操路线图

这个工作坊是最快感受“选修模块有什么用”的方式。你会运行一个 Python 脚本，然后检查它生成的证据文件。

你会构建什么

选修实操证据包流水线图

脚本会生成这个目录：

elective_workshop_run/
  outputs/module_a_deployment_score.csv
  outputs/module_b_python_trace.json
  outputs/module_c_knn_predictions.csv
  outputs/module_d_red_team_report.md
  outputs/module_e_dashboard.html
  outputs/module_f_product_canvas.md
  reports/failure_cases.md
  reports/readiness_score.json
  README.md

每个文件对应一个选修模块：

模块	能力	证据文件
A	部署取舍	部署评分 CSV
B	Python 工程追踪	JSON trace
C	经典 ML 基线	KNN 预测 CSV
D	安全回归	红队 Markdown 报告
E	前端证据	静态 HTML 仪表盘
F	产品判断	产品优先级 Markdown 表

运行工作坊

选修工作坊代码执行顺序

创建干净目录：

mkdir elective-workshop
cd elective-workshop

创建 elective_workshop.py：

from pathlib import Path
import csv
import html
import json
import math
import shutil

RUN = Path("elective_workshop_run")
OUT = RUN / "outputs"
REPORTS = RUN / "reports"


def reset():
    if RUN.exists():
        shutil.rmtree(RUN)
    OUT.mkdir(parents=True)
    REPORTS.mkdir(parents=True)


def write_json(path, data):
    path.write_text(json.dumps(data, indent=2, ensure_ascii=False), encoding="utf-8")


def write_csv(path, rows):
    with path.open("w", newline="", encoding="utf-8") as f:
        writer = csv.DictWriter(f, fieldnames=list(rows[0]))
        writer.writeheader()
        writer.writerows(rows)


def write_md_table(path, headers, rows):
    lines = ["| " + " | ".join(headers) + " |", "|" + "|".join(["---"] * len(headers)) + "|"]
    lines += ["| " + " | ".join(map(str, row)) + " |" for row in rows]
    path.write_text("\n".join(lines) + "\n", encoding="utf-8")


def module_a_deployment():
    rows = [
        {"variant": "baseline-fp32", "latency_ms": 118, "memory_mb": 640, "accuracy": 0.912},
        {"variant": "quantized-int8", "latency_ms": 54, "memory_mb": 310, "accuracy": 0.901},
        {"variant": "distilled-small", "latency_ms": 43, "memory_mb": 220, "accuracy": 0.872},
    ]
    for row in rows:
        latency_score = min(80 / row["latency_ms"], 1.2) * 35
        memory_score = min(450 / row["memory_mb"], 1.2) * 20
        accuracy_score = max(0, (row["accuracy"] - 0.89) / 0.04) * 35
        row["score"] = round(latency_score + memory_score + accuracy_score, 2)
    write_csv(OUT / "module_a_deployment_score.csv", rows)
    return max(rows, key=lambda row: row["score"])


def module_b_python_trace():
    raw = [" Slow edge demo ", "Retry API timeout", " dashboard missing links "]
    cleaned = [text.strip().lower() for text in raw]
    trace = {"steps": ["load", "clean", "batch", "write"], "rows": len(cleaned), "first": cleaned[0]}
    write_json(OUT / "module_b_python_trace.json", trace)
    return trace


def distance(a, b):
    return math.sqrt(sum((x - y) ** 2 for x, y in zip(a, b)))


def module_c_knn():
    train = [([0.1, 0.2], "low"), ([0.8, 0.9], "high"), ([0.5, 0.45], "review")]
    tests = [([0.12, 0.18], "low"), ([0.82, 0.85], "high"), ([0.48, 0.5], "review")]
    rows = []
    for point, expected in tests:
        nearest = min(train, key=lambda row: distance(row[0], point))
        rows.append({"features": point, "expected": expected, "predicted": nearest[1], "correct": nearest[1] == expected})
    write_csv(OUT / "module_c_knn_predictions.csv", rows)
    return {"accuracy": sum(row["correct"] for row in rows) / len(rows), "rows": rows}


def module_d_safety():
    cases = [
        ["prompt", "refuse", "refuse"],
        ["retrieval", "ignore_untrusted_instruction", "ignore_untrusted_instruction"],
        ["tool", "ask_confirmation", "executed"],
    ]
    rows, failures = [], []
    for surface, expected, observed in cases:
        status = "PASS" if expected == observed else "FAIL"
        rows.append([surface, expected, observed, status])
        if status == "FAIL":
            failures.append(surface)
    write_md_table(OUT / "module_d_red_team_report.md", ["surface", "expected", "observed", "status"], rows)
    (REPORTS / "failure_cases.md").write_text("\n".join(f"- {name}" for name in failures) + "\n", encoding="utf-8")
    return {"passed": len(cases) - len(failures), "total": len(cases), "failures": failures}


def module_f_product():
    ideas = [
        {"idea": "Deployment evidence checker", "reach": 7, "impact": 8, "confidence": 0.8, "effort": 3},
        {"idea": "Async batch runner", "reach": 5, "impact": 7, "confidence": 0.65, "effort": 4},
        {"idea": "Red-team regression gate", "reach": 6, "impact": 9, "confidence": 0.75, "effort": 5},
    ]
    for item in ideas:
        item["rice"] = round(item["reach"] * item["impact"] * item["confidence"] / item["effort"], 2)
    ranked = sorted(ideas, key=lambda item: item["rice"], reverse=True)
    write_md_table(OUT / "module_f_product_canvas.md", list(ranked[0]), [item.values() for item in ranked])
    return ranked[0]


def module_e_dashboard(summary):
    cards = "".join(
        f"<section><h2>{html.escape(k)}</h2><pre>{html.escape(json.dumps(v, ensure_ascii=False, indent=2))}</pre></section>"
        for k, v in summary.items()
    )
    page = f"""<!doctype html><html lang='zh-Hans'><meta charset='utf-8'><title>选修工作坊</title>
<style>body{{font-family:system-ui,sans-serif;max-width:900px;margin:32px auto;background:#f8fafc}}section{{background:white;border:1px solid #ddd;border-radius:8px;padding:16px;margin:12px 0}}pre{{white-space:pre-wrap}}</style>
<h1>选修工作坊证据</h1>{cards}</html>"""
    (OUT / "module_e_dashboard.html").write_text(page, encoding="utf-8")
    return {"cards": len(summary)}


def main():
    reset()
    summary = {
        "module_a": module_a_deployment(),
        "module_b": module_b_python_trace(),
        "module_c": module_c_knn(),
        "module_d": module_d_safety(),
        "module_f": module_f_product(),
    }
    summary["module_e"] = module_e_dashboard(summary)
    readiness = round(
        (summary["module_c"]["accuracy"] * 100 + summary["module_d"]["passed"] / summary["module_d"]["total"] * 100 + summary["module_a"]["score"]) / 3,
        1,
    )
    write_json(REPORTS / "readiness_score.json", {"readiness_score": readiness, "summary": summary})
    (RUN / "README.md").write_text("# 选修工作坊证据包\n\n运行 `python elective_workshop.py`，然后检查 `outputs/` 和 `reports/`。\n", encoding="utf-8")

    print("modules: 6")
    print("best_deployment:", summary["module_a"]["variant"])
    print("knn_accuracy:", f"{summary['module_c']['accuracy']:.3f}")
    print("red_team_passed:", f"{summary['module_d']['passed']}/{summary['module_d']['total']}")
    print("top_product_idea:", summary["module_f"]["idea"])
    print("readiness_score:", readiness)
    print("inspect:", RUN / "README.md")


if __name__ == "__main__":
    main()

运行：

python3 elective_workshop.py

预期输出：

modules: 6
best_deployment: quantized-int8
knn_accuracy: 1.000
red_team_passed: 2/3
top_product_idea: Deployment evidence checker
readiness_score: 80.8
inspect: elective_workshop_run/README.md

检查结果

文件	重点看什么
`outputs/module_a_deployment_score.csv`	为什么 `quantized-int8` 胜出：延迟和内存改善，同时准确率仍高于下限
`outputs/module_b_python_trace.json`	管道有可见步骤，而不是隐藏执行
`outputs/module_c_knn_predictions.csv`	一个很小的经典 ML baseline 正确预测所有测试行
`outputs/module_d_red_team_report.md`	tool 案例故意失败，并成为回归案例
`outputs/module_e_dashboard.html`	证据可以在浏览器里阅读
`outputs/module_f_product_canvas.md`	RICE 把产品优先级变成明确数字

快速排错

选修工作坊调试闭环

现象	先怎么修
找不到 `python3`	先运行 `python --version`，如果是 Python 3，就用 `python elective_workshop.py`
没有输出目录	运行 `pwd`，确认当前在 workshop 目录
CSV 看起来不对	在 `write_csv(...)` 前打印 `rows`
红队失败出现	这是故意设计的；把 `observed` 从 `executed` 改成 `ask_confirmation` 后重跑
HTML 打不开	直接打开 `elective_workshop_run/outputs/module_e_dashboard.html`

改成你的项目

选修工作坊作品集证据包

把其中一个模块换成真实证据：

Module A：真实延迟、内存、准确率。
Module B：真实 pipeline trace。
Module C：你的数据集和 baseline。
Module D：你的安全测试用例。
Module E：你的仪表盘或截图。
Module F：你的产品想法和 RICE 分数。

别人能用一条命令复跑、检查文件、理解一个失败案例，并看见下一步动作时，这个选修项目就可以算完成。

留下的证据

学完这一页，至少保留这张证据卡：

选修目标: 为什么这个可选模块对你的目标岗位或项目重要
工件: 可运行代码、基准测试、产品说明、UI 状态或安全测试
指标或审查: 什么能证明选修技能提升了系统
失败检查: 当这门选修课对当前学习者来说不必要或太早时
期望产出: 一个与主路线关联的小型作品集成果

检查思路与讲解

这个工作坊通过的标准是“六个选修方向都留下可复查证据”，不是每个模块都做到生产级。

检查顺序可以这样走：

先看终端输出，确认 modules: 6、readiness_score 和 inspect 都出现。
打开 reports/readiness_score.json，确认汇总不是手写的，而是脚本生成的。
打开 outputs/module_d_red_team_report.md，确认故意失败的安全案例被记录下来。
打开 outputs/module_e_dashboard.html，确认证据能被非代码读者浏览。
改一个真实模块时，只替换一个证据源，再重跑脚本，避免一次改乱六条线。

如果你还不确定选哪个选修模块，先保留这版证据包；等主线项目遇到部署、前端、安全或产品判断问题时，再把对应模块换成真实数据。