From: Sam Smoot <ssmoot@...>
Date: 2007-01-17T13:55:10+09:00
Subject: Re: Can anyone try to solve this problems?


Daniel Martin wrote:
> Why build up a separate hash structure?

It's simpler to my eyes. It's also (admittedly unintuitively) faster
though.

Here's a quick test with your version, my original, and another going
off your idea, but I tried to simplify it a bit:

> #!/opt/local/bin/ruby
> class Symbol
> 	def to_proc
> 	  lambda { |x| x.send(self) }
>   end
> end
>
> module Enumerable
>   def max_by1
>     last_value = nil
>
>     inject do |max_item, item|
>       current_value = yield item
>       if last_value.nil? || current_value > last_value
>         last_value = current_value
>         item
>       else
>         max_item
>       end
>     end
>   end
>
>   def max_by2
>     pairs = inject({}) do |hash, obj|
>       hash[yield obj] = obj; hash
>     end
>     pairs[pairs.keys.max]
>   end
>
>   def max_by3
>     max_value = nil
>     max_item = nil
>
>     each do |item|
>       value = yield(item)
>       max_item, max_value = item, value if max_value.nil? || value > max_value
>     end
>
>     max_item
>   end
> end
>
> a = %w{ one two three four }
>
> 16.times do |i|
>   a = a + a
> end
>
> puts "a.size => #{a.size}"
> require 'benchmark'
>
> Benchmark::bm do |bm|
>   bm.report { puts a.max_by1 &:size }
>   bm.report { puts a.max_by2 &:size }
>   bm.report { puts a.max_by3 &:size }
> end

I couldn't tell you why the 3rd is so much faster than the first. Just
goes to prove the old programming adage though: "Optimizing before
Profiling is a waste of time."

Not that this counts as serious profiling of course. :-)