Muthia,

I got the same results as you, but would add that your test shows even more support for moving out from older systems.

Yes, if someone is stuck with version 8, ordinal may be their best alternative, but definitely not with version 9. Further, if one changes the example to search for the two smallest values, and adds some missing values to the test data, the ordinal version will show the missing values as being the lowest.

Of course that may be useful to some, thus both smallest/largest and ordinal both have their utility (regardless if one can count backwards or not).

Art ------- On Fri, 2 Jan 2009 15:58:19 -0500, Muthia Kachirayan <muthia.kachirayan@GMAIL.COM> wrote:

>On Thu, Jan 1, 2009 at 12:57 PM, Ian Whitlock <iw1sas@gmail.com> wrote: > >> Data _null_ ; >> input Name $ a b c d e f ; >> max1 = ordinal(6,of a--f); >> max2 = ordinal(5,of a--f); >> put name max1 max2 ; >> cards; >> Mahi 12 45 65 87 36 98 >> Suri 32 20 54 86 47 100 >> Nari 05 62 28 65 46 32 >> ;;;; >> >> LARGEST function is not listed in version 8 documentation, but ORDINAL >> has been available for at least 25 years. LARGEST was probably >> introduced for people who cannot count backwards. On the other hand, >> it eliminates the need to know how many variables are involved. >> >> Ian Whitlock >> =============== > > >Ian and all, > >For finding the First Maximum and the Second Maximum I did not choose >ORDINAL() function as I thought that besides background sorting, there may >be other computations to slow the work. After your post I wanted to check my >assumption. Data_NULL brought LARGEST() function then. I had a look at the >speed of the four ways. > >1. Using SORTN() function as I did. > >2. Using other solution where I used Max() function only. > >3. The use of ORDINAL() function as Ian used. > >4. The use of LARGEST() function as Data_NULL_ used. > > > > >The test data : > >data test; >array m aa ab ac ad ae af ag ah ai aj > ba bb bc bd be bf bg bh bi bj > ca cb cc cd ce cf cg ch ci cj > da db dc dd de df dg dh di dj; >do _n_ = 1 to 1000000; > do over m; > m = ceil(ranuni(123) * 999); > end; > output; >end; >run; > >** 1. Using Sortn() ; > >data need; > set test ; > array k[*] _numeric_; > call sortn(of k[*]); > max1 = k[dim(k)]; > max2 = k[dim(k) - 1]; >drop a: b: c: d:; >run; > >**2. Max() is used ; > >In my solution posted earlier, the second Maximum was not correctly caught >in all circumstances and the corrected code is given below. > >data need; >set test ; >array k[*] _numeric_; >do _n_ = 1 to dim(k); > if k[_n_] > max1 then max2 = max1; > else if k[_n_] > max2 then max2 = k[_n_]; > max1 = max(max1,k[_n_]); >end; >drop aa--dj; >run; > >** 3. LARGEST() is used ; > >data need ; >set test ; > max1 = ordinal(40,of aa--dj); > max2 = ordinal(39,of aa--dj); >drop a: b: c: d:; >run; > >** 4. ORDINAL() is used ; > >data need; >set test ; > max1 = largest(1,of aa--dj); > max2 = largest(2,of aa--dj); >drop a: b: c: d:; >run; > >The wall-clock times based on 10 replications are compared and to control >for the CPU-speeds, the relative speeds in time-units are as: > >1. 2.07 >2. 1.88 >3. 1.34 >4. 1.00 > >The SORTN() function (in 9.1 it is experimental ) does not go fast as I >presumed. I do not know whether this has been corrected in 9.2. The >LARGEST() looks best in 9.1. Hence, ORDINAL() seems better alternative in >V8. > >Happy New Year to you and to all SAS-Lers. > >Best regards, > >Muthia Kachirayan