MathGroup Archive 2000

[Date Index] [Thread Index] [Author Index]

Search the Archive

results of pair creation contest

  • To: mathgroup at smc.vnet.net
  • Subject: [mg23374] results of pair creation contest
  • From: Wijnand Schepens <Wijnand.Schepens at rug.ac.be>
  • Date: Fri, 5 May 2000 02:07:07 -0400 (EDT)
  • Organization: RUG
  • Sender: owner-wri-mathgroup at wolfram.com

Hi groupies,

This is an overview of 8 different methods to split a list into a list
of pairs. It is largely the result of the discussions on this mathgroup.
Thanks to everyone who invested time in this.

Example of pair creation:
{a,b,c} -> {{a,b},{a,c},{b,c}}
Every pair must only appear once (i.e. {a,b} is considered to be
equivalent to {b,a}.
The order of the pairs is not very important, as long as the correct
(n-1)/2 pairs are returned.

-------------------------------------------------
8 selected methods
-------------------------------------------------

Trivial implementation using Append:

In[53]:=
divpairs1[lst_List] := Module[{r = {}},    Do[Do[
        r = Append[r, lst[[{j, i}]] ], {j,1, i - 1}], {i, 2,
Length[lst]}]; r]

Using Table:

In[54]:=
divpairs2[lst_List] :=
    Table[Sequence @@ Table[lst[[{j, i}]], {j,1,i - 1}],
{i,2,Length[lst]}]

In[55]:=
divpairs3[lst_List] :=
    (Sequence @@ #1) & /@ Table[lst[[{j,i}]],{i, 2, Length[lst]}, {j, 1,
i - 1}]

In[56]:=
divpairs4[lst_List] :=
  Join @@ Table[lst[[{j, i}]], {i, 2, Length[lst]}, {j, 1, i - 1}]


Mark Harder's method:

In[57]:=
divpairs5[lst_List] :=
  Flatten[Table[lst[[{j, i}]], {i, 2, Length[lst]}, {j, 1, i - 1}], 1]


Using a mask (Carl Woll's method):

In[58]:=
mask[n_] :=
  Module[{r=Range[n - 1]},Flatten[Range @@@ Transpose[{r(n + 1) - (n -
1), r n}]]]

In[59]:=
divpairs6[lst_List] :=
    Module[{m, allpairs},
        m = mask[Length[lst]]; allpairs = Flatten[Outer[List, lst, lst],
1]; allpairs[[m]]
    ]


Hartmut Wolf's method:

In[60]:=
divpairs7[lst_List] :=
    Module[{l = Length[lst]},
        Join@@Array[If[#1 < #2, lst[[{#1, #2}]], Unevaluated[Sequence[]
] ] &,{l, l}]
    ]


Allan Hayes' method, based on an idea of Wagner Truppel (?):

In[61]:=
divpairs8[lst_List] :=
  Transpose[ (Apply[Join, NestList[#, #[lst], Length[lst] - 2] ]) &
      /@ {Drop[#, -1] &, Drop[#, 1] &} ]

-------------------------------------------------
Examples
-------------------------------------------------

All methods produce the same results (possibly in different order) when
applied to a (non-nested) list, e.g.

In[62]:=
divpairs1[{a, b, c, d}]
Out[62]=
{{a, b}, {a, c}, {b, c}, {a, d}, {b, d}, {c, d}}



Note: divpairs6 produces undesired results when applied to a nested
list:
This is what the other methods return (and what we want):
In[70]:=
divpairs1[{{a1, a2}, {b1, b2}, {c1, c2}}]
Out[70]=
{{{a1, a2}, {b1, b2}}, {{a1, a2}, {c1, c2}}, {{b1, b2}, {c1, c2}}}

This is the result of divpairs6 (not what we want):
In[75]:=
divpairs6[{{a1, a2}, {b1, b2}, {c1, c2}}]
Out[75]=
{{{{a2, a1}, {a2, a2}}, {{a2, b1}, {a2, b2}}, {{a2, c1}, {a2, c2}}},
{{{b1,
        a1}, {b1, a2}}, {{b1, b1}, {b1, b2}}, {{b1, c1}, {b1, c2}}},
{{{c2,
        a1}, {c2, a2}}, {{c2, b1}, {c2, b2}}, {{c2, c1}, {c2, c2}}}}
We'll consider divpairs6 in the efficiency-tests anyway


-------------------------------------------------
Efficiency:
-------------------------------------------------

I used my own procedure MultiTiming to time the duration of a
calculation. It first does the calculation two times, neglecting the
timing (which can be abnormally high because of first tries), then it
does n more timings, and returns {min,mean,max} or {best case, mean
case, worst case} of these n timings.

SetAttributes[MultiTiming, HoldAll];
MultiTiming[f_, n_] := Module[{data},
    Timing[f;]; Timing[f;];
    data = Table[Timing[f;][[1, 1]], {n}];
    {Min[data], Statistics`DescriptiveStatistics`Mean[data], Max[data]}]



Now we try all 8 methods on a list of 50 random numbers:

In[78]:=
n = 50; t = Table[Random[], {n}];

In[94]:=
{n, MultiTiming[divpairs1[t], 5], MultiTiming[divpairs2[t], 5],
  MultiTiming[divpairs3[t], 5]
  , MultiTiming[divpairs4[t], 5], MultiTiming[divpairs5[t], 5],
  MultiTiming[divpairs6[t], 5], MultiTiming[divpairs7[t], 5],
  MultiTiming[divpairs8[t], 5]}

Out[94]=
{50, {1.26, 1.274, 1.31}, {0.05, 0.056, 0.06}, {0.05, 0.056, 0.06}, {0.,

    0.044, 0.06}, {0., 0.044, 0.06}, {0., 0.032, 0.06}, {0.11, 0.154,
    0.17}, {0., 0.022, 0.06}}

Apparently the first method divpairs1 is very slow (as expected). We
leave it out for the next timings.
divpairs7 is also somewhat slower than the others.
divpairs8 yields the best average.


Now for n=100 (excluding divpairs1):

In[95]:=
n = 100; t = Table[Random[], {n}];

In[97]:=
{n, MultiTiming[divpairs2[t], 5], MultiTiming[divpairs3[t], 5]
  , MultiTiming[divpairs4[t], 5], MultiTiming[divpairs5[t], 5],
  MultiTiming[divpairs6[t], 5], MultiTiming[divpairs7[t], 5],
  MultiTiming[divpairs8[t], 5]}

Out[97]=
{100, {0.16, 0.198, 0.22}, {0.16, 0.186, 0.22}, {0.16, 0.186, 0.22},
{0.16,
    0.186, 0.22}, {0.11, 0.132, 0.17}, {0.6, 0.614, 0.66}, {0., 0.044,
0.06}}

Results:
divpairs2,3,4 and 5 are rougly equivalent.
divpairs6 is a bit faster
divpairs7 is slow
divpairs8 is clearly the fastest



For n=200 we get (excluding 1)

{200, {0.71, 0.758, 0.83}, {0.76, 0.768, 0.77}, {0.71, 0.736, 0.77},
{0.71,
    0.746, 0.77}, {0.55, 0.55, 0.55}, {2.42, 2.45, 2.47}, {0.05, 0.088,
0.16}}

divpairs8 outbeats divpairs6 (factor 3 worst case, factor 11 best case),
which in turn is better than the others.
Remember also how divpairs6 didn't produce the desired result when
applied to nested lists.

For n=400 we get (excluding 1)

{400, {2.25, 2.79, 4.17}, {3.19, 3.384, 3.46}, {2.96, 3.218, 3.3},
{3.02,
    3.264, 3.35}, {2.25, 2.506, 2.92}, {10.38, 29.668, 51.57}, {0.33,
0.384,
    0.6}}

divpairs8 is on average more than 6 times faster than divpairs6.
For n this large, we see that divpairs1 approaches divpairs6 (at least
in best case)!

That'll be all.

Conclusion: Allan Hayes' method is great. Considering the fact that it
is not obvious at all (I got lots of proposals, of which the 8 cases
above are the best), I suggest this routine be made a part of the
Mathematica system (add-ons...)

Is this the definitive version?

Wijnand Schepens,
University of Ghent, Belgium



  • Prev by Date: Re: Question: Fitting numerical solutions of ord. diff. equ.
  • Next by Date: Re: Random Geometric (long)
  • Previous by thread: Multiple Headers in Complex Data File
  • Next by thread: matrices, dot products and convergents of continued fractions