From: Robert Dober <robert.dober@...>
Date: 2006-04-14T02:05:58+09:00
Subject: Re: map_if, collect_if ???

------=_Part_20206_440501.1144947955072
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline

On 4/13/06, Mauricio Fernandez <mfp@acm.org> wrote:
>
> On Thu, Apr 13, 2006 at 10:53:24PM +0900, Robert Dober wrote:
> > On 4/13/06, Robert Dober <robert.dober@gmail.com> wrote:
> > module Enumerable
> >     def map_if
> >         map{ |x| (r =3D yield(x)) ? r : nil}.compact
> >     end
> > end
> >
> > that one might do as a comprimise between speed, elegance and sanity?
>
> It's quite slow (two blocks in use...).
>
> If you care a bit about performance (but not enough to use C),
>
>   def map_if(&b)
>     a =3D map(&b)
>     a.delete(false) # to get the same semantics as select{|e| e}
>     a.compact!
>     a
>   end
>
> requires half as much mem in the worst case as, and runs ~60% faster than
> map(&b).select{|e| e}:
>
> RUBY_VERSION                                       # =3D> "1.8.4"
> module Enumerable
>   def map_if(&b)
>     a =3D map(&b)
>     a.delete(false) # to get the same semantics as select{|e| e}
>     a.compact!
>     a
>   end
>
>   def map_if2(&b)
>     map(&b).select{|e| e}
>   end
>
>   def map_if3
>     inject([]){|s,x| (v =3D yield(x)) ? s << v : s }
>   end
>
>   def map_if4
>     map{|x| (r =3D yield(x)) ? r : nil}.compact
>   end
> end
>
> require 'benchmark'
>
> TIMES =3D 100
> Benchmark.bmbm(10) do |bm|
>   arr =3D (1..10000).to_a
>   bm.report("compact!"){ TIMES.times{ arr.map_if{true} } }
>   bm.report("select"){ TIMES.times{ arr.map_if2{ true} } }
>   bm.report("inject"){ TIMES.times{ arr.map_if3{ true} } }
>   bm.report("? : test"){ TIMES.times{ arr.map_if4{ true} } }
>
>   # the block wouldn't let us measure the actual performance:
>   #bm.report("compact!"){ TIMES.times{ arr.map_if{|x| x % 7 =3D=3D 0 and =
x} }
> }
>   #bm.report("select"){ TIMES.times{ arr.map_if2{|x| x % 7 =3D=3D 0 and x=
} } }
>   #bm.report("inject"){ TIMES.times{ arr.map_if3{|x| x % 7 =3D=3D 0 and x=
} } }
>   #bm.report("? : test"){ TIMES.times{ arr.map_if4{|x| x % 7 =3D=3D 0 and=
 x} }
> }
> end
>
> # >> Rehearsal ---------------------------------------------
> # >> compact!    0.380000   0.000000   0.380000 (  0.431046)
> # >> select      0.650000   0.000000   0.650000 (  0.725197)
> # >> inject      4.300000   0.020000   4.320000 (  4.647794)
> # >> ? : test    2.400000   0.010000   2.410000 (  2.551757)
> # >> ------------------------------------ total: 7.760000sec
> # >>
> # >>                 user     system      total        real
> # >> compact!    0.390000   0.000000   0.390000 (  0.408975)
> # >> select      0.640000   0.010000   0.650000 (  0.747967)
> # >> inject      3.880000   0.010000   3.890000 (  4.211830)
> # >> ? : test    2.070000   0.000000   2.070000 (  2.199076)
>
>
> --
> Mauricio Fernandez  -   http://eigenclass.org   -  singular Ruby
>
> Very nice work, I just added my original plan to the benchmark, that one
with the side effects
def map_if5(&block)
     find_all(&block).map(&block)
end
it is labeled as "2 yields" below. I spare you the rest of the code, I just
stole from Mauricio, thx ;)
and look at that, it is much faster than I thaught,

but the difference in the relative values also show that these BM are not
really very accurate, I love them though
as they give some general ideas.

                user     system      total        real
compact!    0.330000   0.000000   0.330000 (  0.328067)
select      0.510000   0.000000   0.510000 (  0.525555)
inject      3.160000   0.020000   3.180000 (  3.461044)
? : test    1.590000   0.020000   1.610000 (  1.707898)
2 yields    0.880000   0.000000   0.880000 (  0.897847)

--
Deux choses sont infinies : l'univers et la b=EAtise humaine ; en ce qui
concerne l'univers, je n'en ai pas acquis la certitude absolue.

- Albert Einstein

------=_Part_20206_440501.1144947955072--