Inference apparatus and method for processing instances using linear

Abstract

An inference apparatus predicts the class of a new instance whose class is unknown or generates a rule which is common to an instance set. In the inference apparatus, an instance memory section stores the instance set which each include attribute--value and class. The data type of the attribute--value is discrete type or continuous type. An instance expression section converts the attribute--value of discrete type to a first data expression and the attribute--value of continuous type to a second data expression. A parameter calculation section generates a linear function of each class in which the attribute--value is a variable and calculates a coefficient of each variable in the linear function by assigning one of the first data expression and the second data expression to the variables. A prediction section assigns the attribute--value of the new instance to the variable of the linear function of each class and predicts the class of the new instance by comparing function values among all linear functions of each class. A rule generation section determines coefficients of an orthogonal function by orthogonal expansion of each variable in the linear function, converts each coefficient of the orthogonal function to the first data expression and calculates a logical proposition as the rule from the orthogonal function including the coefficients of the first data expression.