|
I'm sending this on to SAS-L so all can benefit.
______________________________ Forward Header __________________________________
Subject: Re[5]: Is this a bug???
Author: hermans1@westatpo.westat.com at INTERNET
Date: 6/19/96 7:04 PM
I've attached Ian's qualifications of the solution. I believe that
the format will work for larger or smaller numbers so long as you
specify the appropriate format. Ian's solution has more general
applicability (although the put with format method also allows one to
specify user-defined formats as well as default formats). Sig
______________________________________________________________________
Sigurd,
IF round(x,.01)=round(y,.01) THEN .......;
accomplishes the same thing but is a little faster because the format
needs to both round and convert to character. It is also more general
in that it works for numbers over one million and under -100000.
Ian [Whitlock /Westat]
----------------------------------------------------------------------
That is such a simple solution that works beautifully. Anyone know of
any situation where this wouldn't work? It does require that you
specify a level of precision to do your comparison, but for my usage
it would work perfectly. Wow - talk about using the right tool for
the right job. Thanks!
============================================================
== Clifford Aspinall cgaspina@pwinet.upj.com (work) ===
== Kalamazoo, Michigan cgaspina@net-link.net (home) ===
== Disclaimer: Any opinions expressed are not ===
== necessarily mine or my employers ===
============================================================
______________________________ Reply Separator _________________________________
Subject: Re[3]: Is this a bug???
Author: hermans1@WESTATPO.WESTAT.COM at INTERNET
Date: 6/19/96 3:11 PM
I have always wondered about comparisons of floating point numbers in SAS. The
problem seems even worse when we export data from one platform to another. When
in doubt, I compare the converted string values of numbers, as in ...
IF put(x,8.2)=put(y,8.2) THEN .......;
Fortran and C programmers have learned to avoid direct comparisons of FP
numbers. The widespread use of type number variables for storing categorical
codes in SAS datasets makes it more difficult to avoid these questionable forms
of comparisons in SAS programs. Any other ideas? Sig Hermansen
______________________________ Reply Separator _________________________________
Subject: Re[2]: Is this a bug???
Author: WHITLOI1 at REC
Date: 06/19/96 12:50 PM
Subject: Re: Is this a bug???
Summary: Simple problems need simple solutions.
Respondent: Ian Whitlock <whitloi@westat.com>
In response to Andy Yuan's <ANDY_YUAN@SMTPLINK.MSSM.EDU> question why
the test
> DATA TEST1; SET Q1; IF PERCENT EQ 100; RUN; RUN; DATA TEST2; SET Q1;
> IF PERCENT GT 100; RUN;
failed when PERCENT had the value 100 when viewed with the format 20.16
Clifford Aspinall <CGASPINA@AM.PNU.COM> responded with a macro %CMP to
compare real values. Since the answer to real comparisons amounts to
rounding the values compared, I am not sure that a macro is needed, but
if one is to be used it should be simple.
%macro cmp (cond) / parmbuff ;
%let cond = %qsubstr(%quote(&syspbuff),2,%length(&syspbuff)-2) ;
%let cond = %quote(&cond) ;
%unquote(round(%scan(&cond,1,%str( )),1e-12)
%scan(&cond,2,%str( ))
round(%scan(&cond,3,%str( )),1e-12))
%mend cmp ;
The parmbuff option is needed to avoid having the call %CMP ( A = 1 )
result in the error message that the keyword A is not defined. The
macro above assumes an expression of the form EXP OPERAND EXP where
EXP represents an arithmetic expression with no internal blanks and
OPERAND is an arithmetic comparison operator. The quoting functions
are used to handle people who make tests like
data _null_;
or = 3 ;
ne = 1.7+.3 ;
if %cmp ( or eq ne+1 ) then put 'ok' ;
run ;
Ian Whitlock
| 
