Calculate a mean average from a Ruby array

Ruby doesn’t provide a native method to generate an average (mean) value from an array of integers. Its built-in Math library focuses on more complex calculations and there’s no built-in #average or #mean method on Array.

I'm co-chairing RailsConf 2024 in Detroit May 7–9.
Come and join us →

This leaves us free to create our own implementation, introducing the opportunity to shoot ourselves in the foot.

Ensure you use…

…Array#sum when calculating the mean average from an array of integers:

a = [1, 2, 3, 4, 5, 6, 7, 8]

a.sum(0.0) / a.size
#=> 4.5

Why?

Using the #sum method from Array is many, many times faster than using the alternative, inject.

The #sum method was added to Array in Ruby 2.4, which is why you might see alternative implementations in other places on the Internet.

In order to compare the performance of the different implementations we can use the benchmark-ips gem.

require "benchmark/ips"

# Generate a 1,000 element array
a = Array.new(1000) { |_| rand(1000) }

Benchmark.ips do |x|
  x.report("sum(0.0) / size") do
    a.sum(0.0) / a.size
  end
  x.report('inject(0.0) / size') do
    a.inject(0.0) { |result, i| result + i } / a.size
  end
  x.compare!
end

The results (on my laptop) show a 50× speed improvement when using the native #sum. This method was built into the language to provide exactly this sort of performance improvement.

Calculating -------------------------------------
   sum(0.0) / size  680.425k (± 6.5%) i/s -  3.432M in 5.06s
inject(0.0) / size   13.513k (± 5.0%) i/s - 67.586k in 5.01s

Comparison:
   sum(0.0) / size: 680425.2 i/s
inject(0.0) / size:  13512.7 i/s - 50.35x  slower

Although there are variations of using #sum and then #size that perform similarly, the biggest performance win is choosing to use the native #sum method in the first place.

The benchmarking code for a wider variety of implementations:

require "benchmark/ips"

# Generate a 1,000 element array
a = Array.new(1000) { |_| rand(1000) }

Benchmark.ips do |x|
  x.report("sum(0.0) / size") do
    a.sum(0.0) / a.size
  end
  x.report("sum.to_f / size") do
    a.sum.to_f / a.size
  end
  x.report("sum / size.to_f") do
    a.sum / a.size.to_f
  end
  x.report("sum.fdiv(size)") do
    a.sum.fdiv(a.size)
  end
  x.report('inject(0.0, :+) / size') do
    a.inject(0.0, :+) / a.size
  end
  x.report('inject(0.0) / size') do
    a.inject(0.0) { |result, i| result + i } / a.size
  end
  x.report('inject(0).to_f / size') do
    a.inject(0) { |result, i| result + i }.to_f / a.size
  end
  x.report('inject(0) / size.to_f') do
    a.inject(0) { |result, i| result + i } / a.size.to_f
  end
  x.report('inject(0).fdiv(size)') do
    a.inject(0) { |result, i| result + i }.fdiv(a.size)
  end
  x.compare!
end

…and the comparison results:

Comparison:
       sum(0.0) / size: 668222.4 i/s
       sum / size.to_f: 660291.4 i/s - same-ish
       sum.to_f / size: 655929.1 i/s - same-ish
        sum.fdiv(size): 621960.0 i/s - same-ish
inject(0.0, :+) / size:  30823.6 i/s - 21.68x slower
 inject(0) / size.to_f:  18740.7 i/s - 35.66x slower
 inject(0).to_f / size:  18320.1 i/s - 36.47x slower
  inject(0).fdiv(size):  18082.6 i/s - 36.95x slower
    inject(0.0) / size:  15264.0 i/s - 43.78x slower

Anything else?

Rails’ Active Record does include an #average method. It’s used to perform calculations on numerical columns in the database directly in SQL.

If that is your specific use case, you should definitely use that method. Instantiating Active Record models and subsequently iterating over the Array in Ruby will nearly always be slower.

Still running UK’s friendliest, Ruby event on Friday 28th June.
Ice cream + Ruby →