SAS KNN - Extract Distance Measure

macworthy — Mon, 11 Sep 2017 17:35:11 GMT

I am runing a KNN algorithm and would like to be able to extract the distance for each tested variable. For example, if the data is trained on Dataset A and tested on Dataset B, then what is the distance between a given record in Dataset B and its nearest neighbor in Dataset A. I am using the PROC DISCRIM procedure to execute this.

DATA MASTER;
	INFILE DATALINES;
	INPUT k  errorrate;
	DATALINES;
RUN;

%macro KNND;
	%do i=1 %TO 50;
		%IF %SYSFUNC(MOD(&i,2))=1 %THEN %DO;

			PROC DISCRIM 
				DATA=TRAIN 
				TESTDATA=TEST
				TESTOUT=_score&i
				METHOD=NPAR
				K=&i;
				CLASS BAD;
				VAR VAR1 VAR2 VAR3 VAR4; 
			RUN; 

				PROC SQL noprint;
					CREATE TABLE stage&i AS 
						SELECT &i AS k,
						SUM(CASE WHEN BAD = _INTO_ THEN 0 ELSE 1 END)/COUNT(NIIN) AS errorrate
				FROM _score&i;
			QUIT;

			PROC APPEND BASE=MASTER DATA=stage&i;
			RUN; 


			%END;


%END; 

%MEND KNND;

%KNND

Re: SAS KNN - Extract Distance Measure

PGStats — Mon, 11 Sep 2017 18:01:58 GMT

You could concatenate datasetA and datasetB and extract a sub-matrix of the matrix provided by proc distance.

topic Re: SAS KNN - Extract Distance Measure in Statistical Procedures

SAS KNN - Extract Distance Measure

Re: SAS KNN - Extract Distance Measure