From: Karsten Meier <developer@...>
Date: 2012-01-23T23:08:32+09:00
Subject: Re: uniq with count; better way?

Ok, I tried some benchmarks. We have now even more variables, as they
also depend on "maxval" from the dataset.

maxval = 1000
ar = [].tap{|a| 1_000_000.times {a << rand(maxval)}}

b.report("Meier:") {
  n.times {
    hist =  Array.new(maxval+1, 0)
    ar.each{|x| hist[x] += 1;}
    result = Hash.new(0)
    0.upto(maxval){|i| result[i] = hist[i] unless hist[i] == 0}
    result
  }
}



On my jruby and my windows- mri 1.8.7 my algorithm was fastest for
maxvalue of 10, 100 or 10000, for example:

SIZE
1000000
MAXVAL
10000
                     user     system      total        real
Ralph Shneiver:  0.533000   0.000000   0.533000 (  0.518000)
Meier:           0.312000   0.000000   0.312000 (  0.312000)
Keinich #1       0.814000   0.000000   0.814000 (  0.814000)

(I have no 1.9.3 yet on my windows PC, so it may be different there)

But here are two observations:
1) speed-up is not so big as I expected. In C, I expect array lookup to
be factors better than hash calculation (followed by an array-lookup in
the hash-table...). In Ruby it seem to be not so much faster. But the
speedup gets bigger for bigger values of maxval.

2) My algorithm sometimes run much much slower when #kennich1 had run
before mine. :-( It seem to get worse with big values of maxval, but not
with jruby --1.9 option. It is not the array-allocate itself that is the
problem.
Is it possible that group_by changes the internal array structure, So I
get a non-continous-array?

Regards

Karsten Meier

-- 
Posted via http://www.ruby-forum.com/.