Hi, I happened to have a question on the prox measure in rf: I guess the splitting in "tree growing" when running in an unsupervised mode has to be totally random (i.e. mtry=1 and splitting value is randomly selected) but I am not sure if the guessing is correct. But if the class label is available, should rf grow trees in a classification way so that it can give a better measure on prox? thanks for your comments, Weiwei -- Weiwei Shi, Ph.D "Did you always know?" "No, I did not. But I believed..." ---Matrix III [[alternative HTML version deleted]]