Jay, Model based approaches (see www.latentclass.com) such as that used in the Latent GOLD program provide parameter estimates that can be applied to new cases. As the formula for computing the posteriors may be complicated in some cases, a trick that you can use with the Latent GOLD program to have the program output posterior probabilities for (new) cases not used in the model estimation is to 1) append the new cases to the bottom of your data file and 2) create a 'case weight' variable that equals 1 for your original cases, and a small constant such as 0.0000000001 (say 1E-100) for each new case. Then estimate the same model again, using this case weight as a weight and requesting classification output to a file. You will get the classification information for the new records in addition to the observed records.
Jason ----- Original Message ----- From: "Jay Liu" <[EMAIL PROTECTED]> To: <[email protected]> Sent: Tuesday, July 26, 2005 1:37 PM Subject: Prediction in clustering > Dear all, > > > > Apart from how to determine the number of clusters, another difficulty > > in clustering (I think) is how to predict cluster memberships of new > > data. This is very straight forward in classification but I can't think > > of a single clustering method I know can do this. I guess some > > model-based techniques maybe can do this but frankly, I have no clue at all. > > > > > > > Jay. > >
