Natural data typically contains outliers or even faulty values. They may correspond to atypical cases or be a sign of a broken measurement device. For better performance such values should be filtered out from the data, if possible. Since the SOM can handle partial data, there is no need to discard the whole vector, however. In ENTIRE the scaling is done by specifying the bounds inside which the value of a given component must be, and then comparing each data vector with them. If the component value is outside the allowed bounds, it is masked out.