Hi, there: This is a question a little bit off topic but I believe many people using bioconductor might have this situation so I ask it here and hope I can get some suggestion. I have a result which looks like this: net num.genes overall.error overall.pred.error 1 custom 5 0.15625 0.05263 The overall.error is (b+c)/(a+b+c+d) from cross-validation for training data; while the overall.pred.error is the one for test data. Since the sample sizes of training and test data are different, it gives me the result which performs better in test than training. I am wondering if there are some other metrics to evalute this classification error rate so that it can consider the effects of sample size. thanks -- Weiwei Shi, Ph.D Research Scientist GeneGO, Inc. "Did you always know?" "No, I did not. But I believed..." ---Matrix III

Shouldn't there be error bars (i.e., confidence intervals) around those error estimates, that ought to get smaller when you use more samples? Best, Kevin Weiwei Shi wrote: > Hi, there: > > This is a question a little bit off topic but I believe many people > using bioconductor might have this situation so I ask it here and hope > I can get some suggestion. > > I have a result which looks like this: > net num.genes overall.error overall.pred.error > 1 custom 5 0.15625 0.05263 > > The overall.error is (b+c)/(a+b+c+d) from cross-validation for > training data; while the overall.pred.error is the one for test data. > Since the sample sizes of training and test data are different, it > gives me the result which performs better in test than training. I am > wondering if there are some other metrics to evalute this > classification error rate so that it can consider the effects of > sample size. > > thanks >