Path: utzoo!utgpu!water!ljdickey
From: ljdickey@water.waterloo.edu (Lee Dickey)
Newsgroups: comp.lang.apl
Subject: Re: MODE
Message-ID: <1451@water.waterloo.edu>
Date: 8 Mar 88 13:56:57 GMT
References: <1428@water.waterloo.edu> <49700001@uicsrd.csrd.uiuc.edu>
Reply-To: ljdickey@water.waterloo.edu (Lee Dickey)
Organization: U. of Waterloo, Ontario
Lines: 30

In article <49700001@uicsrd.csrd.uiuc.edu> jaxon@uicsrd.csrd.uiuc.edu writes:
> ...
>For a large array, the outer product in the above solution
>can be slow.  A solution that makes fewer comparisons is:
> ...

I think that Greg is right, as is David Lauer, who observed that

	..."+/NUB jot.= X" can produce an
	immensely large intermediate array...

David's solution, quite similar to Greg's is :

 Z <- MODE X
 X <- X[gradeup X]            # SORT X
 I <- X not_eq 1 shift X      # FIND BOUNDARIES BETWEEN GROUPS OF LIKE NUMBERS
 I <- ( neg 1 drop I),1       # DEAL WITH EDGE CONDITIONS
 NUB <- I/X                   # FIND UNIQUE NUMBERS
 H <- I/ iota rho I           # 
 H <- H-0, neg 1 drop H       # FIND LENGTH OF RUN (IE: FREQUENCY)
 I <- H iota (max /H)         # FIND MOST FREQUENT (SMALLEST IF SEVERAL)
 Z <- NUB[I]                  # LOOK-UP MOST FREQUENT X VALUE

David's address is: David Laur <DMLAUR@PUCC.bitnet>

-- 
 L. J. Dickey, Faculty of Mathematics, University of Waterloo.
	ljdickey@waterloo.edu
	ljdickey@watdcs.BITNET
	ljdickey@water.UUCP	...!uunet!water!ljdickey