Add SVM lesson: Hard-margin vs Soft-margin with interactive demo

[LMK89]

Summary

Added a comprehensive Vietnamese-language educational lesson on Support Vector Machines (SVM), focusing on the practical differences between Hard-margin and Soft-margin approaches with real-world credit risk classification examples.

Key Changes

• New lesson file: 2026-08-31-svm-support-vector-machine-28b46348.html (643 lines)

  • Complete HTML5 lesson with embedded CSS and JavaScript
  • Dark/light theme toggle with localStorage persistence
  • Responsive design with CSS Grid and Flexbox

• Content structure:

  • Quick warmup quiz (3 questions) to assess prior knowledge
  • Concise SVM definition with mathematical notation
  • Real-world scenario: bank credit risk scoring (1D classification)
  • Four detailed state comparisons showing SVM behavior:
    • Hard-margin with clean data (ideal case)
    • Hard-margin with outliers (infeasible/fails)
    • Soft-margin with large C (overfits to noise)
    • Soft-margin with reasonable C (robust solution)
  • Comparative summary table
  • Interactive demo: visualize threshold selection with/without outliers
  • Data quality guidelines (good vs bad practices)
  • Self-check quiz (3 questions with instant feedback)

• Interactive features:

  • Toggle outlier inclusion in demo dataset
  • Radio buttons to switch between Hard-margin and Soft-margin modes
  • Animated number line visualization showing data points and decision threshold
  • Step-by-step solver output with color-coded feedback
  • Warmup and quiz scoring with localStorage support

• Updated README: Added entry linking to the new SVM lesson

Notable Implementation Details

• Uses 1D (univariate) classification for manual calculation accessibility
• Demonstrates why Hard-margin fails on real data through concrete example (B0 outlier)
• Emphasizes the C parameter's role in balancing margin width vs. training error tolerance
• All styling uses CSS custom properties for theme switching
• JavaScript animations (fadeInDown, slideUp, slideDown, pulse) for progressive disclosure
• No external dependencies — pure HTML/CSS/JavaScript

https://claude.ai/code/session_011NfjZZebBNSnmoi6X4RMN1