An Information Measure for Classification

摘要: 1. The class to which each thing belongs. 2. average properties of class. 3. deviations from the its parent If things are found be concentrated in a small area region measurement space then will small, and with reference most information about is given by naming it In this case may recorded much more briefly than if classification had not been used. We suggest that best results briefest recording all attribute information. context, we regard measurements as being message thing. Shannon (1948) showed where messages regarded nominating occurrence particular event among universe possible events, needed record series such minimised encoded so length proportional minus logarithm relative frequency nominates. required greatest when frequencies equal. here nominate positions 5 1 points representing attributes things. expected density everywhere uniform, cannot simple list measured values. However, markedly non-uniform, application