Minitest::Benchmark

Subclass Benchmark to create your own benchmark runs. Methods starting with “bench_” get executed on a per-class.

Public Class Methods

bench_exp(min, max, base = 10) click to toggle source

Returns a set of ranges stepped exponentially from min to max by powers of base. Eg:

bench_exp(2, 16, 2) # => [2, 4, 8, 16]

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 35
def self.bench_exp min, max, base = 10
  min = (Math.log10(min) / Math.log10(base)).to_i
  max = (Math.log10(max) / Math.log10(base)).to_i

  (min..max).map { |m| base ** m }.to_a
end

bench_linear(min, max, step = 10) click to toggle source

Returns a set of ranges stepped linearly from min to max by step. Eg:

bench_linear(20, 40, 10) # => [20, 30, 40]

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 48
def self.bench_linear min, max, step = 10
  (min..max).step(step).to_a
rescue LocalJumpError # 1.8.6
  r = []; (min..max).step(step) { |n| r << n }; r
end

bench_range() click to toggle source

Specifies the ranges used for benchmarking for that class. Defaults to exponential growth from 1 to 10k by powers of 10. Override if you need different ranges for your benchmarks.

See also: ::bench_exp and ::bench_linear.

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 61
def self.bench_range
  bench_exp 1, 10_000
end

Public Instance Methods

assert_performance(validation, &work) click to toggle source

Runs the given work, gathering the times of each run. Range and times are then passed to a given validation proc. Outputs the benchmark name and times in tab-separated format, making it easy to paste into a spreadsheet for graphing or further analysis.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  validation = proc { |x, y| ... }
  assert_performance validation do |n|
    @obj.algorithm(n)
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 83
def assert_performance validation, &work
  range = self.class.bench_range

  io.print "#{self.name}"

  times = []

  range.each do |x|
    GC.start
    t0 = Minitest.clock_time
    instance_exec(x, &work)
    t = Minitest.clock_time - t0

    io.print "\t%9.6f" % t
    times << t
  end
  io.puts

  validation[range, times]
end

assert_performance_constant(threshold = 0.99, &work) click to toggle source

Runs the given work and asserts that the times gathered fit to match a constant rate (eg, linear slope == 0) within a given threshold. Note: because we’re testing for a slope of 0, R^2 is not a good determining factor for the fit, so the threshold is applied against the slope itself. As such, you probably want to tighten it from the default.

See www.graphpad.com/guides/prism/8/curve-fitting/reg_intepretingnonlinr2.htm for more details.

Fit is calculated by fit_linear.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  assert_performance_constant 0.9999 do |n|
    @obj.algorithm(n)
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 127
def assert_performance_constant threshold = 0.99, &work
  validation = proc do |range, times|
    a, b, rr = fit_linear range, times
    assert_in_delta 0, b, 1 - threshold
    [a, b, rr]
  end

  assert_performance validation, &work
end

assert_performance_exponential(threshold = 0.99, &work) click to toggle source

Runs the given work and asserts that the times gathered fit to match a exponential curve within a given error threshold.

Fit is calculated by fit_exponential.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  assert_performance_exponential 0.9999 do |n|
    @obj.algorithm(n)
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 153
def assert_performance_exponential threshold = 0.99, &work
  assert_performance validation_for_fit(:exponential, threshold), &work
end

assert_performance_linear(threshold = 0.99, &work) click to toggle source

Runs the given work and asserts that the times gathered fit to match a straight line within a given error threshold.

Fit is calculated by fit_linear.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  assert_performance_linear 0.9999 do |n|
    @obj.algorithm(n)
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 193
def assert_performance_linear threshold = 0.99, &work
  assert_performance validation_for_fit(:linear, threshold), &work
end

assert_performance_logarithmic(threshold = 0.99, &work) click to toggle source

Runs the given work and asserts that the times gathered fit to match a logarithmic curve within a given error threshold.

Fit is calculated by fit_logarithmic.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  assert_performance_logarithmic 0.9999 do |n|
    @obj.algorithm(n)
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 173
def assert_performance_logarithmic threshold = 0.99, &work
  assert_performance validation_for_fit(:logarithmic, threshold), &work
end

assert_performance_power(threshold = 0.99, &work) click to toggle source

Runs the given work and asserts that the times gathered curve fit to match a power curve within a given error threshold.

Fit is calculated by fit_power.

Ranges are specified by ::bench_range.

Eg:

def bench_algorithm
  assert_performance_power 0.9999 do |x|
    @obj.algorithm
  end
end

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 213
def assert_performance_power threshold = 0.99, &work
  assert_performance validation_for_fit(:power, threshold), &work
end

fit_error(xys) click to toggle source

Takes an array of x/y pairs and calculates the general R^2 value.

See: en.wikipedia.org/wiki/Coefficient_of_determination

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 222
def fit_error xys
  y_bar  = sigma(xys) { |_, y| y } / xys.size.to_f
  ss_tot = sigma(xys) { |_, y| (y    - y_bar) ** 2 }
  ss_err = sigma(xys) { |x, y| (yield(x) - y) ** 2 }

  1 - (ss_err / ss_tot)
end

fit_exponential(xs, ys) click to toggle source

To fit a functional form: y = ae^(bx).

Takes x and y values and returns [a, b, r^2].

See: mathworld.wolfram.com/LeastSquaresFittingExponential.html

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 237
def fit_exponential xs, ys
  n     = xs.size
  xys   = xs.zip(ys)
  sxlny = sigma(xys) { |x, y| x * Math.log(y) }
  slny  = sigma(xys) { |_, y| Math.log(y)     }
  sx2   = sigma(xys) { |x, _| x * x           }
  sx    = sigma xs

  c = n * sx2 - sx ** 2
  a = (slny * sx2 - sx * sxlny) / c
  b = ( n * sxlny - sx * slny ) / c

  return Math.exp(a), b, fit_error(xys) { |x| Math.exp(a + b * x) }
end

fit_linear(xs, ys) click to toggle source

Fits the functional form: a + bx.

Takes x and y values and returns [a, b, r^2].

See: mathworld.wolfram.com/LeastSquaresFitting.html

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 281
def fit_linear xs, ys
  n   = xs.size
  xys = xs.zip(ys)
  sx  = sigma xs
  sy  = sigma ys
  sx2 = sigma(xs)  { |x|   x ** 2 }
  sxy = sigma(xys) { |x, y| x * y  }

  c = n * sx2 - sx**2
  a = (sy * sx2 - sx * sxy) / c
  b = ( n * sxy - sx * sy ) / c

  return a, b, fit_error(xys) { |x| a + b * x }
end

fit_logarithmic(xs, ys) click to toggle source

To fit a functional form: y = a + b*ln(x).

Takes x and y values and returns [a, b, r^2].

See: mathworld.wolfram.com/LeastSquaresFittingLogarithmic.html

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 259
def fit_logarithmic xs, ys
  n     = xs.size
  xys   = xs.zip(ys)
  slnx2 = sigma(xys) { |x, _| Math.log(x) ** 2 }
  slnx  = sigma(xys) { |x, _| Math.log(x)      }
  sylnx = sigma(xys) { |x, y| y * Math.log(x)  }
  sy    = sigma(xys) { |_, y| y                }

  c = n * slnx2 - slnx ** 2
  b = ( n * sylnx - sy * slnx ) / c
  a = (sy - b * slnx) / n

  return a, b, fit_error(xys) { |x| a + b * Math.log(x) }
end

fit_power(xs, ys) click to toggle source

To fit a functional form: y = ax^b.

Takes x and y values and returns [a, b, r^2].

See: mathworld.wolfram.com/LeastSquaresFittingPowerLaw.html

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 303
def fit_power xs, ys
  n       = xs.size
  xys     = xs.zip(ys)
  slnxlny = sigma(xys) { |x, y| Math.log(x) * Math.log(y) }
  slnx    = sigma(xs)  { |x   | Math.log(x)               }
  slny    = sigma(ys)  { |   y| Math.log(y)               }
  slnx2   = sigma(xs)  { |x   | Math.log(x) ** 2          }

  b = (n * slnxlny - slnx * slny) / (n * slnx2 - slnx ** 2)
  a = (slny - b * slnx) / n

  return Math.exp(a), b, fit_error(xys) { |x| (Math.exp(a) * (x ** b)) }
end

sigma(enum, &block) click to toggle source

Enumerates over enum mapping block if given, returning the sum of the result. Eg:

sigma([1, 2, 3])                # => 1 + 2 + 3 => 6
sigma([1, 2, 3]) { |n| n ** 2 } # => 1 + 4 + 9 => 14

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 324
def sigma enum, &block
  enum = enum.map(&block) if block
  enum.inject { |sum, n| sum + n }
end

validation_for_fit(msg, threshold) click to toggle source

Returns a proc that calls the specified fit method and asserts that the error is within a tolerable threshold.

 
               # File minitest-5.14.2/lib/minitest/benchmark.rb, line 333
def validation_for_fit msg, threshold
  proc do |range, times|
    a, b, rr = send "fit_#{msg}", range, times
    assert_operator rr, :>=, threshold
    [a, b, rr]
  end
end

Home Classes Methods

In Files

Parent

Methods

Files

Class/Module Index

Minitest::Benchmark

Public Class Methods

Public Instance Methods