MathLab/skills/math-textbook-compiler/references/workflow.md

Textbook Compiler Workflow

Overview

This skill converts mathematical textbooks into interactive, hands-on courseware with a hacker mindset: treat formulas as legacy code to refactor.

Source Textbooks

Located at /mnt/ai/textbooks/:

1. lihang-code - Statistical Learning Methods (李航)

   • Phase 1: Symbol desensitization
   • Topics: Convex optimization, curse of dimensionality

2. d2l-zh - Dive into Deep Learning

   • Phase 2: Deep dive
   • Topics: Computational graphs, automatic differentiation, tensor mappings

3. Book-Mathematical-Foundation-of-Reinforcement-Learning

   • Phase 3: RL Math Foundations
   • Topics: State spaces, MDPs, dynamic agent interaction

Compilation Pipeline

Step 1: Parse Source

python3 scripts/compile-textbook.py --source /mnt/ai/textbooks/d2l-zh --output staging/

Step 2: Generate Courseware (6 Modules)

For each day's content:

1. Learning Objectives - What you'll learn
2. Core Concepts - Key ideas with visual intuition
3. Symbol Decoder - Translate math to code
4. The Math - Core derivations (strict LaTeX)
5. Optimization Bottlenecks - Engineering considerations
6. OJ Mission - Hands-on coding exercise

Step 3: Create Exercises

• Use templates/exercise_template.py
• Implement from scratch with numpy/torch
• No high-level APIs (torch.nn forbidden)

Step 4: Generate Tests

• Use templates/test_template.py
• Shape assertions
• Numerical correctness checks with np.testing.assert_allclose

Cron Schedule

Job	Schedule	Action
weekend-batch-compile	Sat 10:00 AM	Generate 7 days of materials
daily-deploy	Daily 14:00 PM	Deploy + git push + notify

Output Structure

staging/
├── course_day1.html
├── course_day2.html
├── ...
├── exercises/
│   ├── day1_task.py
│   ├── day2_task.py
│   └── ...
└── tests/
    ├── test_day1.py
    ├── test_day2.py
    └── ...

Deployment

Daily at 14:00 PM:

1. Move from staging/ to courseware/, exercises/, tests/
2. Git commit and push
3. Send Telegram notification