1. Validation data is necessary for tuning and slecting your model, but testing dataset is optional. If you have enough dataset, you want to test(verify) your selected model , you can use test data. Assessment is based on validation dataset, but for some algorithm such as KNN, assessment is based on training.
2. I dont know which version you are using. In 5.2, EM scores both training and validation dataset, and validation is the default for model selection.
[此贴子已经被作者于2008-5-21 4:44:21编辑过]