Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments

sasowiczka123 — Sat, 06 Jun 2020 18:08:54 GMT

Hello, I'm trying to make banks clients segmentation and I failed in making some plots. Not working properly code is bold. Please help me.

preview of standrazied datasetdistribution of RMSE - wrongAlso not correctSuper wrong cluster plot

proc stdize data=dane out=Stand method=std;
var Duration Age Credit_amount;
run;



/* k-means*/
proc fastclus data=Stand out=Clust
maxclusters=5 maxiter=100;
var Duration Age Credit_amount;
run;

proc freq data=Clust;
tables Purpose*Cluster;
run;



/*Definition of a macro program called multseed, which takes the number of initial measures as the argument.  It allows to find several cluster solutions with different starting points*/
Title "Multseed macro for finding multiple 
cluster solutions with different starting seeds";
%macro multseed(n);
%do i = 1 %to &n;
proc surveyselect data=Clust out=seeds n=5
seed=&i noprint;
run;

proc fastclus data=Clust maxc=5 converge=0
drift seeds=seeds outstat=out least=2
noprint mean=meanout;
var Duration Age Credit_amount;
run;

data out;
set out (where=(_type_="WITHIN_STD"));
seed=&i;
run;

data ans;
set out
%if &i>1 %then %do; 
ans
%end;
;
keep seed over_all Duration Age Credit_amount;
run;
%end;
%mend;


/*Using multseed with n=100*/
%multseed(100);




/*Comparative analysis of results for 100 segmentation models from different starting seeds*/
proc univariate data=ans plot;
var OVER_ALL;
id seed;
run;


/* it's not working */
/*Building a histogram of RMSE values for 100 different solutions of  the K-means method at different starting points*/
Title "Distribution of RMSE";
proc sgplot data=ans;
histogram over_all / scale=count binwidth=.01;
xaxis display=(nolabel);
run;



/*Building a scatter plot of estimated cluster centers for 100 different solutions of the K-means method at various start points*/
Title "Estimated cluster centers of 100 different K-means solutions";
proc sgplot data=ans;
scatter x=Duration y=Credit_amount;
run;

/*Best solution out of 100 seeds*/ 
Title "Best solution out of 100 seeds";
proc surveyselect data=Stand out=seeds n=5 seed=66;
run;
proc fastclus data=Stand maxc=5 converge=0 drift seed=seeds
outstat=out least=2 mean=meanout out=asgn;
var Duration Age Credit_amount;
run;


/* THIS IS NOT WORKING PROPERLY */

/*Creating a chart showing average clusters and assignments to them as part of the best solution*/
Title "Plot with the best means and cluster assignments";
data clusout;
set asgn meanout 
(keep=Duration Age Credit_amount);
length clus $6.;
if cluster=. then clus="Center";
else clus=put(cluster, $1.);
label clus="Cluster";
run;
data myanno;
drawspace='datavalue'; function='line';
x1=-.1; y1=3.5; x2=4.5; y2=3.5; output;
x1=4.5; y1=4.5; x2=7.1; y2=4.5; output;
run;
proc sgplot data=clusout sganno=myanno;
scatter x=Duration y=Credit_amount / group=clus;
refline 4.5 / axis=x;
lineparm x=2.5 y=0 slope=-1;
run;

Re: Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments

ChrisNZ — Fri, 12 Jun 2020 06:50:35 GMT

Moved to the procedures forum.

Re: Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments

SteveDenham — Fri, 12 Jun 2020 17:10:17 GMT

It is very difficult to identify an error from the statement "This is not working properly." What errors, warnings and notes are in the log? Is the plot that is produced somehow not what was expected? If that is the case, tell us what is expected (or at least is wanted). With these pieces of information, we are in a much better place for providing tips/guidance/other code that may be useful.

SteveDenham

topic Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments in Statistical Procedures

Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments

Re: Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments

Re: Proc fastclus, proc sgplot how to create a plot with the best means and cluster assignments