Data process unit and data process unit control program

A computer-readable storage medium containing instructions for controlling a data process unit to perform a method, the method comprising:

• classifying a plurality of predetermined data on a plurality of objects into a plurality of groups based on a plurality of specific conditions;
• generating a plurality of pattern models which have 4-dimensional or higher dimensional elements for each group of predetermined data based on the predetermined data classified by the classifying step;
• calculating a mathematical distance among the pattern models generated by the pattern model generating step of the respective groups;
• converting the plurality of pattern models into the same number of low dimensional vectors corresponding to the pattern models while maintaining distance relationship among the pattern models, based on the mathematical distance calculated by the mathematical distance calculating step;
• and
• displaying the plurality of low dimensional vectors corresponding to pattern models as coordinate points in a low dimensional space of the same dimension as the low dimensional vectors corresponding to pattern models while maintaining the distance relationship, based on values of low dimensional elements.

A computer-readable storage medium containing instructions for controlling a pattern model search unit to perform a method, the method comprising:

• automatically dividing coordinate points displayed in a low dimensional space by a data process unit into a plurality of regions in the low dimensional space;
• generating pattern models for each region based on predetermined data corresponding to coordinate points of the low dimensional vectors corresponding to pattern models contained in the segment regions;
• acquiring predetermined data on a new object;
• and
• calculating a likelihood of regional pattern models corresponding to respective segment regions in relation to the acquired predetermined data and searching for a regional pattern model with recognition performance suitable for recognizing the predetermined data on the new object based on the calculated likelihood.

A data process unit comprising:

• data classification means for classifying a plurality of predetermined data on a plurality of objects into a plurality of groups based on a plurality of specific conditions;
• pattern model generating means for generating a plurality of pattern models which have 4-dimensional or higher dimensional elements for each group of predetermined data based on the predetermined data classified by the data classification means;
• mathematical distance calculating means for calculating mathematical distance among the pattern models generated by the pattern model generating means for the respective groups;
• pattern model converting means for converting the plurality of pattern models into the same number of low dimensional vectors corresponding to the pattern models in the lower dimension while maintaining distance relationship among the pattern models, based on the mathematical distance calculated by the mathematical distance calculating means;
• and
• low dimensional vector corresponding to pattern model display means for displaying the plurality of low dimensional vectors corresponding to pattern models as coordinate points in a low dimensional space of the same dimension as the low dimensional vectors corresponding to pattern models while maintaining the distance relationship, based on values of low dimensional elements, wherein
• the mathematical distance calculating means uses the occurrence frequency of each pattern unit in the plurality of predetermined data on the plurality of objects when calculating the mathematical distance.

A data process unit comprising:

• data classification means for classifying a plurality of predetermined data on a plurality of objects into a plurality of groups based on a plurality of specific conditions;
• pattern model generating means for generating a plurality of pattern models which have 4-dimensional or higher dimensional elements for each group of predetermined data based on the predetermined data classified by the data classification means;
• mathematical distance calculating means for calculating mathematical distance among the pattern models generated by the pattern model generating means for the respective groups;
• performance suitable for recognizing the predetermined data on the new object based on the calculated likelihood.

A computer-readable storage medium containing instructions for controlling a data process unit to perform a method, the method comprising:

• classifying a plurality of predetermined data on a plurality of objects into a plurality of groups based on a plurality of specific conditions;
• generating a plurality of pattern models which have 4-dimensional or higher dimensional elements for each group of predetermined data based on the predetermined data classified by the data classifying step;
• calculating a mathematical distance among the pattern models generated by the pattern model generating step of the respective groups;
• converting the plurality of pattern models into the same number of low dimensional vectors corresponding to the pattern models in the lower dimension while maintaining a distance relationship among the pattern models, based on the mathematical distance calculated by the mathematical distance calculating step;
• and
• displaying the plurality of low dimensional vectors corresponding to pattern models as coordinate points in a low dimensional space of the same dimension as the low dimensional vectors corresponding to pattern models while maintaining the distance relationship, based on values of low dimensional elements,
• wherein the mathematical distance calculating step uses an occurrence frequency of each pattern unit in the plurality of predetermined data on the plurality of objects when calculating the mathematical distance.

A computer-readable storage medium containing instructions for controlling a data process unit to perform a method, the method comprising:

• classifying a plurality of predetermined data on a plurality of objects into a plurality of groups based on a plurality of specific conditions;
• generating a plurality of pattern models which have 4-dimensional or higher dimensional elements for each group of predetermined data based on the predetermined data classified by the data classifying step;
• calculating a mathematical distance among the pattern models generated by the pattern model generating step of the respective groups;
• converting the plurality of pattern models into the same number of low dimensional vectors corresponding to the pattern models in the lower dimension while maintaining a distance relationship among the pattern models, based on the mathematical distance calculated by the mathematical distance calculating step;
• displaying the plurality of low dimensional vectors corresponding to pattern models as coordinate points in a low dimensional space of the same dimension as the low dimensional vectors corresponding to pattern models while maintaining the distance relationship, based on values of low dimensional elements;
• automatically dividing the coordinate points of the plurality of low dimensional vectors corresponding to pattern models displayed in the low dimensional space in the vector displaying step into a plurality of regions in the low dimensional space;
• generating pattern models for each region based on predetermined data corresponding to coordinate points of the low dimensional vectors corresponding to pattern models contained in the segment regions;
• acquiring predetermined data on a new object;
• and
• calculating likelihood of regional pattern models for respective segment regions in relation to the acquired predetermined data and searching for a regional pattern model with recognition performance suitable for recognizing the predetermined data on the new object based on the calculated likelihood.