RE: Re: Friendly Challenge 2: sort by column

*To*: mathgroup at smc.vnet.net*Subject*: [mg35079] RE: [mg35067] Re: Friendly Challenge 2: sort by column*From*: "DrBob" <majort at cox-internet.com>*Date*: Mon, 24 Jun 2002 03:21:07 -0400 (EDT)*Reply-to*: <drbob at bigfoot.com>*Sender*: owner-wri-mathgroup at wolfram.com

>>Merely the behavior of the program when there are identical items in the sort column, possibly resulting in ambiguity. In general I prefer the Rotate solution as it more closely follows the default behavior of Sort. I don't know why you call that the default behavior; Sort preserves existing order in case of ties, and there's no option not to. (Changing the ordering function can ELIMINATE the ties, but it doesn't change Sort's behavior in CASE of ties.) That seems odd to me, too, since a sort that ignored existing order could be faster. Bobby Treat -----Original Message----- From: "Mr. Wizard" [mailto:gleam at flashmail.com] To: mathgroup at smc.vnet.net Subject: [mg35079] [mg35067] Re: Friendly Challenge 2: sort by column MathGroup readers and challenge participants, Now that I have let a few days go by, allowing time for those who would choose to participate to write and post their methods, I will share the function that I have been using. First, for Hartmut Wolf, who wrote: Paul, what do you mean "handling duplicates" exactly? Merely the behavior of the program when there are identical items in the sort column, possibly resulting in ambiguity. In general I prefer the Rotate solution as it more closely follows the default behavior of Sort. Of course, this is slower however. <snip> ad[[Ordering[RotateLeft[ad,{0,2}]]]]//MatrixForm And there you have it! That's all that I was after. :-) Next, to DrBob, who wrote: Another interpretation (the one I prefer) is to preserve preexisting order in case of ties. That allows us to sort by one column, then by another, and get the result we'd expect. After some experimentation, I found that hw4 and hw5 (and hence Ordering itself) preserve order already, without modification. As you will see, this is included in my function with the following syntax: positionSort[a, n] uses Rotate positionSort[a, {n}] sorts by only column n, preserving other ordering. positionSort[a, {n, o, ...} sorts secondarily by o, etc. n can be negative to give the column n from the end. A third argument can be given to specify an arbitrary sort function. Here is a variation of the function that I use. The original code uses pattern matching differently (in place of the If statement, for example), and is too verbose for posting, though it yields a small improvement in performance: Notebook[{ Cell["Array test function", "Text"], Cell[BoxData[ \(mtest = #~TensorRank~2 > 1 &\)], "Input"], Cell["Optional sort function handling", "Text"], Cell[BoxData[ \(zz = HoldPattern[ z : \(\((_Symbol | _Function | _Sequence)\) : Sequence[]\)]\)], "Input"], Cell["Handling of trivial parameters", "Text"], Cell[BoxData[{ \(positionSort[a_, 0 | {0}] := a\), "\n", \(positionSort[a_List, 1] := Sort@a\)}], "Input"], Cell["\<\ Sort by only one column (disregard the rest); arbitrary sort function\ \ \>", "Text"], Cell[BoxData[ \(positionSort[a_?mtest, {p_Integer} | \(p : {_Integer .. }\), zz] := a\[LeftDoubleBracket] a\[LeftDoubleBracket]All, p\[RightDoubleBracket]~Ordering~ z\[RightDoubleBracket]\)], "Input"], Cell["Arbitrary sort function", "Text"], Cell[BoxData[ \(positionSort[a_?mtest, p_Integer, zz] := a\[LeftDoubleBracket]a~RotateLeft~{0, If[p < 0, p, p - 1]}~ Ordering~z\[RightDoubleBracket]\)], "Input"] }] Paul