Class: MoreMath::LinearRegression

Inherits:
Object
  • Object
show all
Defined in:
lib/more_math/linear_regression.rb

Overview

This class computes a linear regression for the given image and domain data sets.

Linear regression is a statistical method that models the relationship between a dependent variable (image) and one or more independent variables (domain). It fits a linear equation to observed data points to make predictions or understand relationships.

The implementation uses the least squares method to find the best-fit line y = ax + b, where ‘a’ is the slope and ‘b’ is the y-intercept.

Examples:

Basic usage

# Create a linear regression from data points
image_data = [2, 4, 6, 8, 10]
domain_data = [1, 2, 3, 4, 5]
lr = LinearRegression.new(image_data, domain_data)

# Access the fitted line parameters
puts lr.a  # slope
puts lr.b  # y-intercept

# Make predictions
predicted_y = lr.a * 6 + lr.b  # Predict y for x=6

Statistical analysis

# Check if the slope is significantly different from zero
lr.slope_zero?(0.05)  # Returns true if slope is not statistically significant

# Calculate coefficient of determination (R²)
puts lr.r2  # R-squared value indicating model fit

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(image, domain = (0...image.size).to_a) ⇒ LinearRegression

Creates a new LinearRegression instance with image and domain data.

Initializes the linear regression model using the provided data points. The domain data represents independent variables (x-values) and the image data represents dependent variables (y-values).

Examples:

Creating a linear regression

image = [1, 2, 3, 4, 5]
domain = [0, 1, 2, 3, 4]
lr = LinearRegression.new(image, domain)

Parameters:

  • image (Array<Numeric>) —

    Array of dependent variable values (y-coordinates)

  • domain (Array<Numeric>) (defaults to: (0...image.size).to_a) —

    Array of independent variable values (x-coordinates)

Raises:

  • (ArgumentError) —

    If image and domain arrays have unequal sizes



46
47
48
49
50
51
 
# File 'lib/more_math/linear_regression.rb', line 46

def initialize(image, domain = (0...image.size).to_a)
  image.size != domain.size and raise ArgumentError,
    "image and domain have unequal sizes"
  @image, @domain = image, domain
  compute
end

Instance Attribute Details

#a ⇒ Float (readonly)

The slope of the line.

Returns the calculated slope (a) of the best-fit line y = ax + b.

Returns:

  • (Float) —

    The slope coefficient of the linear regression



72
73
74
 
# File 'lib/more_math/linear_regression.rb', line 72

def a
  @a
end

#b ⇒ Float (readonly)

The offset of the line.

Returns the calculated y-intercept (b) of the best-fit line y = ax + b.

Returns:

  • (Float) —

    The y-intercept coefficient of the linear regression



79
80
81
 
# File 'lib/more_math/linear_regression.rb', line 79

def b
  @b
end

#domain ⇒ Array<Numeric> (readonly)

The domain data as an array.

Returns the independent variable values used in the regression.

Returns:

  • (Array<Numeric>) —

    Array of x-values from the original data



65
66
67
 
# File 'lib/more_math/linear_regression.rb', line 65

def domain
  @domain
end

#image ⇒ Array<Numeric> (readonly)

The image data as an array.

Returns the dependent variable values used in the regression.

Returns:

  • (Array<Numeric>) —

    Array of y-values from the original data



58
59
60
 
# File 'lib/more_math/linear_regression.rb', line 58

def image
  @image
end

Instance Method Details

#r2 ⇒ Float

Returns the coefficient of determination (R²).

R² measures the proportion of the variance in the dependent variable that is predictable from the independent variable(s). It ranges from 0 to 1, where higher values indicate better fit.

Examples:

Checking model fit

lr = LinearRegression.new([1, 2, 3], [0, 1, 2])
puts lr.r2  # 1.0 for perfect linear relationship

Returns:

  • (Float) —

    The R-squared value (0.0 to 1.0)



133
134
135
136
137
138
139
140
 
# File 'lib/more_math/linear_regression.rb', line 133

def r2
  image_seq = MoreMath::Sequence.new(@image)
  sum_res   = residuals.inject(0.0) { |s, r| s + r ** 2 }
  [
    1.0 -  sum_res / image_seq.sum_of_squares,
    0.0,
  ].max
end

#residuals ⇒ Array<Float>

Returns the residuals of this linear regression.

Residuals are the differences between observed values and predicted values from the regression line. They represent the error in prediction for each data point.

Examples:

Calculating residuals

lr = LinearRegression.new([1, 2, 3], [0, 1, 2])
puts lr.residuals  # [0.0, 0.0, 0.0] for perfect fit

Returns:

  • (Array<Float>) —

    Array of residual values (observed - predicted)



115
116
117
118
119
120
121
 
# File 'lib/more_math/linear_regression.rb', line 115

def residuals
  result = []
  @domain.zip(@image) do |x, y|
    result << y - (@a * x + @b)
  end
  result
end

#slope_zero?(alpha = 0.05) ⇒ Boolean

Checks if the slope is significantly different from zero.

Performs a t-test to determine whether the slope coefficient is statistically significant at the given significance level (alpha). This test helps determine if there’s a meaningful linear relationship between the independent and dependent variables.

Examples:

Testing slope significance

lr = LinearRegression.new([1, 2, 3, 4, 5], [2, 4, 6, 8, 10])
lr.slope_zero?  # => false (slope is significantly different from zero)
lr.slope_zero?(0.1)  # => false (still significant at 10% level)

Parameters:

  • alpha (Float) (defaults to: 0.05) —

    The significance level (default: 0.05, or 5%)

Returns:

  • (Boolean) —

    true if the slope is not significantly different from zero, false otherwise

Raises:

  • (ArgumentError) —

    If alpha is not in the range 0..1



96
97
98
99
100
101
102
103
 
# File 'lib/more_math/linear_regression.rb', line 96

def slope_zero?(alpha = 0.05)
  (0..1) === alpha or raise ArgumentError, 'alpha should be in 0..100'
  df = @image.size - 2
  return true if df <= 0 # not enough values to check
  t = tvalue
  td = TDistribution.new df
  t.abs <= td.inverse_probability(1 - alpha.abs / 2.0).abs
end