jtomschroeder / crystalline Goto Github PK

So I tried to benchmark the priority queue against my own queue with the following code to benchmark the code in this repo, and I find the results I get pretty strange.

require "benchmark"

require "./priority_queue"

values = 100_000.times.map { rand(100_000_000) }.to_a

Benchmark.bm do |x|
  queue = PriorityQueue(Int32).new
  x.report("crystalline - push") {
    values.each do |v|
      queue.push(v, v)
    end
  }
  x.report("crystalline - pop") {
    values.size.times do |v|
      queue.pop
    end
  }
  puts queue.size

  queue2 = PriorityQueue(Int32).new
  x.report("crystalline - mixed") {
    values.each_with_index do |v, i|
      queue2.push(v, v)
      if i % 4 == 0
        queue2.pop
      end
    end
  }
  puts queue2.size

end

This outputs

linus@donk:~/prog/pairing_heap$ crystal run src/benchmark.cr --release --no-debug
0
0
                          user     system      total        real
crystalline - push    55.120000   1.290000   56.410000 (  36.032381)
crystalline - pop     0.440000   0.000000   0.440000 (  0.261351)
crystalline - mixed   0.120000   0.000000   0.120000 (  0.069474)

The disconnect between push and mixed is pretty staggering.

For the record, my queue is at https://github.com/yxhuvud/pairing_heap/blob/master/src/priority_queue.cr . It is faster but has slightly different semantics(it picks the least number for starters) and other method names, but it could be interesting to compare against.

While I can lock this library to a particular SHA in my shard.yml, it'd be nice to have a shard.yml with a version so that there can be some semblance of semver throughout the project lifetime.

Thanks for making a great collection!

A>

I am in need of an Ordered Set and am planning to implement one from this library implementation of Red Black tree.

As such to avoid code duplication, I will reimplement the RB Tree Map as an RB Set of pairs. What do you think of it?