topic Gaussian mixture model proc fmm in Statistical Procedures

Gaussian mixture model proc fmm

yuhhwa — Tue, 06 Nov 2018 12:02:01 GMT

in case of multiple variables(+100 variables),the following statement is OK?

proc fmm data=XXX;
class XXX;←think separately about model?
model A=,B=C=,・・・・・・ = / k=X;
run;

Re: Gaussian mixture model proc fmm

Rick_SAS — Wed, 07 Nov 2018 13:12:41 GMT

Your statements are not valid syntax.

1. How does the classification variable enter the models?

2. The K= option requires a constant value, such as K=2 or K=3.

3. You cannot use commas to separate models. You cannot put multiple models on one MODEL statement.

It is not clear what you are trying to accomplish, but here is a guess.

A. It looks like you might have many response variables named Y1, Y2, Y3, ..., Y100

B. It looks like you are trying to fit them all with a mixture of k Gaussian distributions.

If those assumptions are correct then you should convert your data from wide to long format. Let Y be the name of the long variable that contains the values for variables Y1, ..., Y100 and let B be an indicator variable that indicates which observations correspond to which variables. Then you can use syntax such as the following:

proc fmm data=LongData plots=none;
   by B;
   model Y= / k=2;
run;

Almost surely you do not want 100 pages of output, so you should suppress ODS output and write the results (parameter estimates?) to a data set. You should study the techniques in the following articles:

Simulation in SAS: The slow way or the BY way

Turn off ODS when running simulations in SAS

An easy way to run thousands of regressions in SAS

Re: Gaussian mixture model proc fmm

yuhhwa — Thu, 08 Nov 2018 01:16:43 GMT

thank you for replying.

>>It is not clear what you are trying to accomplish, but here is a guess.

A. It looks like you might have many response variables named Y1, Y2, Y3, ..., Y100

B. It looks like you are trying to fit them all with a mixture of k Gaussian distributions.

A and B is right, but I want to cluster combination of multiple variables as well as k-means.

   proc fastclus data=stan out=clust maxclusters=3;
      var v1 v2 v3;
   run;

I think the way you taught me is to cluster variables separately. Is it right?

if you are OK,please tell me how shoud I wright the statement Gaussian mixture model compared with k-means.

thank you in advance.