A non-parametric diagnostic for exploring the relation between pairs of sampled data
Donald R. Hush, Chaouki Tanios Abdallah · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 1993
This report describes a non-parametric method for exploring the functional relation between pairs of sampled data. Specifically, a diagnostic tool is developed to help categorize the relation into one of three groups: one-to-one (OTO), many-to-one (MTO), or one-to-many (OTM). This categorization has strong implications for learning, since it is impossible to learn a (deterministic) relation that is OTM. The tool can be used not only to detect OTM relations, but also to correct for this situation by editing the data so that samples contributing to the OTM behavior are removed. Once this is accomplished, learning can be carried out with the edited data set. In addition, this diagnostic can provide an indication of how much {open_quotes}noise{close_quotes} is present in the data, or more generally, how much variation there is in the data that cannot be described by a deterministic model. That is, it can be used to determine limits on how accurately one can expect to learn the data relation, regardless of the learning model used.