Module: Musa::Darwin

Included in:
All
Defined in:
lib/musa-dsl/generative/darwin.rb

Overview

Evolutionary selection algorithm based on fitness evaluation.

Darwin implements a selection algorithm inspired by natural selection, evaluating and ranking a population of objects based on defined measures (features and dimensions) and weights. Objects are measured, scored, and sorted by fitness for evolutionary algorithms or optimization.

Core Concepts

  • Population: Collection of objects to evaluate
  • Measures: Evaluation criteria for each object
    • Features: Boolean flags (present/absent)
    • Dimensions: Numeric values (continuous)
    • Die: Mark object as non-viable (eliminated)
  • Weights: Importance multipliers for features/dimensions
  • Fitness: Calculated score from normalized dimensions and features
  • Selection: Population sorted by fitness (highest first)

Evaluation Process

  1. Measure: Evaluate each object with measures block
  2. Normalize: Scale dimension values to 0-1 range
  3. Weight: Apply weights to dimensions and features
  4. Score: Calculate total fitness for each object
  5. Sort: Order population by fitness (descending)

Musical Applications

  • Select best harmonic progressions from generated candidates
  • Rank melodic variations by aesthetic criteria
  • Optimize rhythmic patterns for complexity/simplicity
  • Evolve musical structures through iterative selection

A musical selection reads the same way, with the piece's own helpers doing the judging (a reading: has_parallel_fifths? and friends are yours, not the framework's):

darwin = Musa::Darwin::Darwin.new do
  measures do |progression|
    die if has_parallel_fifths?(progression)

    dimension :voice_leading, -total_voice_leading_distance(progression)

    feature :authentic_cadence if ends_with_V_I?(progression)
    feature :plagal_cadence if ends_with_IV_I?(progression)

    dimension :chromaticism, -count_chromatic_notes(progression)
  end

  weight voice_leading: 3.0,
         authentic_cadence: 2.0,
         plagal_cadence: 1.0,
         chromaticism: 1.5
end

best = darwin.select(candidates).first(10)

Examples:

Basic selection with features and dimensions

darwin = Musa::Darwin::Darwin.new do
  measures do |object|
    # Kill objects with unwanted property
    die if object[:bad_property]

    # Binary features
    feature :has_alpha if object[:type] == :alpha
    feature :has_beta if object[:type] == :beta

    # Numeric dimension (negative to prefer lower values)
    dimension :complexity, -object[:complexity].to_f
  end

  # Weight contributions to fitness
  weight complexity: 2.0, has_alpha: 1.0, has_beta: -0.5
end

population = [{ type: :alpha, complexity: 1 },
              { type: :beta,  complexity: 3 },
              { type: :alpha, complexity: 5 },
              { type: :alpha, complexity: 2, bad_property: true }]

darwin.select(population).collect { |o| [o[:type], o[:complexity]] }
# => [[:alpha, 1], [:alpha, 5], [:beta, 3]]
# Sorted by fitness, and the one that died is gone rather than last.

See Also:

Defined Under Namespace

Classes: Darwin