What is the fastest way to sort a Hash?

Question

People often ask what is the best way to sort a hash, but then they don\'t ask the needed follow-up question about what is the fastest way, which really determ

Answer 1

This is a comparison of sort and sort_by when accessing a more complex object:

require 'fruity'

RUBY_VERSION # => "2.2.2"

class Foo
  attr_reader :key
  def initialize(k)
    @key = k
  end

  def <=>(b)
    self.key <=> b.key
  end
end

HASH = Hash[[*(1..100)].shuffle.map{ |k| [Foo.new(k), 1] }]
compare do
  _sort1 { HASH.sort.to_h }
  _sort_by { HASH.sort_by{ |k,v| k.key }.to_h }
end
# >> Running each test 32 times. Test will take about 1 second.
# >> _sort_by is faster than _sort1 by 2.7x ± 0.1

Answer 2

What is the fastest way to sort a Hash?

require 'fruity'

HASH = Hash[('a'..'z').to_a.shuffle.map{ |k| [k, 1] }]

def sort_hash1(h)
  h.sort.to_h
end

def sort_hash2(h)
  Hash[h.sort]
end

def sort_hash3(h)
  Hash[h.sort_by{ |k, v| k }]
end

def sort_keys(h)
  keys = h.keys.sort
  Hash[keys.zip(h.values_at(*keys))]
end

puts "Running on Ruby v#{ RUBY_VERSION }"
puts

compare do
  do_sort_hash1 { sort_hash1(HASH) } if [].respond_to?(:to_h)
  do_sort_hash2 { sort_hash2(HASH) }
  do_sort_hash3 { sort_hash3(HASH) }
  do_sort_keys { sort_keys(HASH) }
end

Running the above code on a Mac OS laptop results in the following output:

# >> Running on Ruby v2.2.2
# >> 
# >> Running each test 256 times. Test will take about 1 second.
# >> do_sort_keys is faster than do_sort_hash3 by 39.99999999999999% ± 10.0%
# >> do_sort_hash3 is faster than do_sort_hash1 by 1.9x ± 0.1
# >> do_sort_hash1 is similar to do_sort_hash2

And:

# >> Running on Ruby v1.9.3
# >> 
# >> Running each test 256 times. Test will take about 1 second.
# >> do_sort_keys is faster than do_sort_hash3 by 19.999999999999996% ± 10.0%
# >> do_sort_hash3 is faster than do_sort_hash2 by 4x ± 0.1

---

Doubling the hash size:

HASH = Hash[[*('a'..'z'), *('A'..'Z')].shuffle.map{ |k| [k, 1] }]

Results in:

# >> Running on Ruby v2.2.2
# >> 
# >> Running each test 128 times. Test will take about 1 second.
# >> do_sort_keys is faster than do_sort_hash3 by 50.0% ± 10.0%
# >> do_sort_hash3 is faster than do_sort_hash1 by 2.2x ± 0.1
# >> do_sort_hash1 is similar to do_sort_hash2

And:

# >> Running on Ruby v1.9.3
# >> 
# >> Running each test 128 times. Test will take about 1 second.
# >> do_sort_keys is faster than do_sort_hash3 by 30.000000000000004% ± 10.0%
# >> do_sort_hash3 is faster than do_sort_hash2 by 4x ± 0.1

---

The values will change depending on the hardware, but the relative results shouldn't change.

Fruity was chosen over using the built-in Benchmark class for simplicity.

This was prompted by "Sort hash by key, return hash in Ruby".

Answer 3

Here are some more interesting things to consider:

require 'fruity'

puts "Running Ruby v#{ RUBY_VERSION }"
# >> Running Ruby v2.2.2

require 'fruity'

puts "Running Ruby v#{ RUBY_VERSION }"
# >> Running Ruby v2.2.2

This looks at the differences using an integer as a key:

HASH = Hash[[*(1..100)].shuffle.map{ |k| [k, 1] }]
compare do
  _sort1 { HASH.sort.to_h }
  _sort2 { HASH.sort{ |a, b| a[0] <=> b[0] }.to_h }
  _sort3 { HASH.sort{ |a, b| a.first <=> b.first }.to_h }
  _sort_by { HASH.sort_by{ |k,v| k }.to_h }
end
# >> Running each test 64 times. Test will take about 1 second.
# >> _sort_by is faster than _sort2 by 70.0% ± 1.0%
# >> _sort2 is faster than _sort3 by 19.999999999999996% ± 1.0%
# >> _sort3 is faster than _sort1 by 19.999999999999996% ± 1.0%

This looks at the differences using a single-character string as the key:

HASH = Hash[[*('a'..'Z')].shuffle.map{ |k| [k, 1] }]
compare do
  _sort1 { HASH.sort.to_h }
  _sort2 { HASH.sort{ |a, b| a[0] <=> b[0] }.to_h }
  _sort3 { HASH.sort{ |a, b| a.first <=> b.first }.to_h }
  _sort_by { HASH.sort_by{ |k,v| k }.to_h }
end
# >> Running each test 16384 times. Test will take about 1 second.
# >> _sort1 is similar to _sort3
# >> _sort3 is similar to _sort2
# >> _sort2 is faster than _sort_by by 1.9x ± 0.1