DataLab — High-Precision Extrapolation, Fitting & Error Propagation

中文 · English — 一套基于 mpmath 高精度计算的科研工具,支持桌面 GUI 和 Web 双前端。 A research-grade scientific toolkit with high-precision (mpmath) sequence extrapolation, curve fitting, error propagation, and weighted statistics — shipped as both a PySide6 desktop app and a Flask web app sharing one core.

---

🇨🇳 中文文档

✨ 主要功能

模式	用途	算法
序列外推	把数值方法在不同步长下的近似值外推到极限	Richardson、Wynn-ε(Shanks)、Levin u-transform、自定义公式
曲线拟合	带不确定度的非线性最小二乘拟合 + 自动模型选择(AIC/BIC)	多项式、Padé、自定义表达式;可选 MCMC 后验优化
误差传递	对 1.23(4)[-2] 形式数据按公式自动传播 1σ 不确定度	SymPy 偏导数 + Monte Carlo
统计平均	加权 / 算术平均 + 标准误差 + RMS 散度	单边 / 双边、加权方差

🚀 一键安装(开箱即用)

平台	下载	大小
🍎 macOS (Apple Silicon)	DataLab-macos.zip	~64 MB
🪟 Windows 10/11 (x64)	DataLab-windows.zip	~83 MB

包内已嵌入 Python 解释器 + PySide6 + 全部科学计算依赖,无需额外安装任何环境。

macOS 使用方法

1. 解压 DataLab-macos.zip → DataLab.app
2. 双击运行(首次启动右键 → "打开" 绕过 Gatekeeper 警告)
3. 可选:把 DataLab.app 拖到 /Applications/ 注册到启动台

Windows 使用方法

1. 解压 DataLab-windows.zip → DataLab\ 目录
2. 双击 DataLab\DataLab.exe
3. 不要单独移动 .exe — Python 运行时在同目录的 _internal\ 下

📷 界面截图

截图位于 docs/screenshots/,来自 macOS 上运行的桌面 GUI(Qt 内部 grab 直出,1400×900)。

主界面:左侧计算面板,右侧 5 个标签页(数据 / 结果 / 日志 / LaTeX / PDF 预览)

数据表:列宽均分、加减行列按钮、文本视图切换、CSV 粘贴

拟合结果:参数 + 不确定度 Markdown 表 + χ²/AIC/BIC + 可选 MCMC 后验

LaTeX 标签页:行号 + 引擎下拉框 + 字体大小;支持 pdflatex / xelatex / Tectonic

🏃 快速上手(用内置示例)

examples/ 目录已附 5 个示例文件,直接加载即可学习每种模式的数据格式:

examples/
├── extrapolation_richardson.txt    # 外推:3 列序列 → Richardson 三点加速
├── fitting_powerlaw.txt            # 拟合:y = 2.5x^-0.7 + 0.1 + 高斯噪声
├── error_propagation.txt           # 误差传递:V1, V2 + 各自 σ
├── statistics_weighted.txt         # 统计:10 次加权测量
└── constants.txt                   # 物理常数:ALPHA, G, c, h, hbar, kB, NA, e

步骤:

1. 启动 GUI → 顶部 mode 下拉框选模式(如 "Extrapolation")
2. "打开数据文件" → 选 examples/<对应文件>.txt
3. 误差传递模式:在"常数文件"字段加载 examples/constants.txt,公式输入 (V1+V2)/V1
4. 点 "开始执行" → 看 "结果" 标签
5. 切到 "LaTeX" 标签 → 选引擎(默认 Tectonic,首次会自动下载 ~30 MB)→ 点编译 → "PDF 预览" 标签看渲染结果

🔢 不确定度括号语法

DataLab 在数据文件 / 公式输入 / 常数文件里都支持紧凑的括号不确定度:

写法	数值	σ
1.234(5)	1.234	0.005(末位 5 个单位)
1.234(5)[-3]	1.234e-3	0.005e-3
7.2973525693(11)[-3]	7.2973525693e-3	1.1e-12
299792458	299792458	0(精确值)

📦 LaTeX 引擎

• Tectonic(默认推荐) — 单文件二进制,GUI 首次使用会自动下载到 ~/.datalab/bin/,自动联网解析缺失的 LaTeX 包。无需 TeX Live。
• pdflatex / xelatex — 如果你已装了 TeX Live / MiKTeX,从下拉框选这两个即可。

🔬 高精度计算

• 默认精度:50–80 dps(可在精度 spinbox 调整,上限 100 万 dps)
• 所有计算包在 precision_guard() 上下文里,跨 worker 线程隔离 mp.dps
• 自动单元测试覆盖 Mathematica 高精度参考值(tests/test_*_mathematica_reference.py)

🛠️ 从源码运行(开发者)

# 桌面 GUI
pip install -r gui_requirements.txt
python data_extrapolation_gui.py

# Web 应用
pip install -r web_requirements.txt
python app_web/server.py    # 默认 http://127.0.0.1:8000

# 测试
pip install -r requirements-test.txt
QT_QPA_PLATFORM=offscreen pytest -q   # 770+ tests

详细架构 + 跨平台打包指南见 docs/ARCHITECTURE.md 和 docs/desktop/ / docs/web/。

---

🇬🇧 English

✨ Features

Mode	Purpose	Methods
Extrapolation	Take a numerical method's approximations at decreasing step sizes and extrapolate to the limit	Richardson, Wynn-ε (Shanks), Levin u-transform, custom formulas
Curve Fitting	Non-linear least-squares with measurement uncertainties + automatic model selection (AIC/BIC)	Polynomial, Padé, arbitrary expression; optional MCMC posterior refinement
Error Propagation	Propagate 1σ uncertainties from 1.23(4)[-2]-style data through user formulas	SymPy partial derivatives + Monte Carlo cross-check
Statistics	Weighted / arithmetic mean + standard error + RMS scatter	One-sided / two-sided, weighted variance

🚀 One-click install (zero dependencies)

Platform	Download	Size
🍎 macOS (Apple Silicon)	DataLab-macos.zip	~64 MB
🪟 Windows 10/11 (x64)	DataLab-windows.zip	~83 MB

The bundle ships Python + PySide6 + the full scientific stack — no extra runtime install needed.

macOS

1. Unzip DataLab-macos.zip → DataLab.app
2. Double-click. First launch: Right-click → "Open" to bypass the Gatekeeper notice.
3. Optional: drag DataLab.app into /Applications/ to register it.

Windows

1. Unzip DataLab-windows.zip → DataLab\ folder
2. Double-click DataLab\DataLab.exe
3. Don't move .exe out of the folder — the Python runtime lives in the sibling _internal\ directory.

📷 Screenshots

See docs/screenshots/ — captured directly from the desktop GUI on macOS via Qt's in-process QWidget.grab() (1400×900 window).

🏃 Quick start (with bundled examples)

The examples/ directory ships five working input files so a new user can load real data into each mode without inventing test data:

examples/
├── extrapolation_richardson.txt   # 3-col Richardson three-point sequence
├── fitting_powerlaw.txt           # y = 2.5 x^-0.7 + 0.1 + Gaussian noise
├── error_propagation.txt          # V1, V2 each with measurement σ
├── statistics_weighted.txt        # 10 independent weighted measurements
└── constants.txt                  # ALPHA, G, c, h, hbar, kB, NA, e (with σ)

Workflow: start the GUI → pick a mode → load the matching example file → "Run" → switch to the "LaTeX" tab → pick an engine (Tectonic is the default and auto-downloads on first use) → compile → see PDF in the preview tab.

🔢 Uncertainty notation

DataLab accepts the compact parenthetical notation used by the high-precision data community:

Token	Value	σ
1.234(5)	1.234	0.005 (last-digit units)
1.234(5)[-3]	1.234e-3	0.005e-3
299792458	299792458	0 (exact)

📦 LaTeX engines

• Tectonic (default, recommended) — single ~30 MB binary, auto-downloads to ~/.datalab/bin/ on first use, fetches missing LaTeX packages over the net per-document. No TeX Live required.
• pdflatex / xelatex — pick from the dropdown if you already have TeX Live or MiKTeX installed.

🛠️ Running from source

pip install -r gui_requirements.txt
python data_extrapolation_gui.py

# or web app:
pip install -r web_requirements.txt
python app_web/server.py

# tests:
pip install -r requirements-test.txt
QT_QPA_PLATFORM=offscreen pytest -q   # 770+ tests

For architecture + per-platform packaging see docs/ARCHITECTURE.md and docs/desktop/ / docs/web/.

---

🤝 Reporting issues / Contributing

Issues and pull requests welcome at GitHub Issues. When reporting a numerical bug please include the exact input data (or attach the offending file) and the active precision setting — high-precision regressions are usually trivial to reproduce given the inputs.

For development setup, coding conventions, and the PR flow see CONTRIBUTING.md. Security vulnerabilities go through the private channel described in SECURITY.md, not public issues. Per-version changes are tracked in CHANGELOG.md.

📜 License

DataLab is released under the MIT License. The bundled LaTeX engine (Tectonic) is downloaded on-demand and licensed separately under the MIT License.