Regression tree generation method and apparatus therefor
AssigneeInternational Business Machines Corporation
InventorsYasuhiko Morimoto, Hiromu Ishii, Shinichi Morishita
Life of this patent
Estimating the term…
Abstract
To generate a more preferable regression tree by using two predicative numeric attributes as a rule for splitting. For N.times.M buckets of a plane that has two axes corresponding to two predicative numeric attributes of data in a database, which belongs to one node of a regression tree, the number u(i, j) of data, belonging to a bucket (coordinate (i, j)), and sum v(i, j) of an objective numeric attribute value of data belonging to the bucket are stored. A bucket region which maximizes the interclass variance is segmented from the plane. Nodes are prepared for the data included in the extracted bucket region and for data outside the bucket region.
Claims18
A method for generating a regression tree for objective numeric attribute of data in a database, the method comprising:
- forming a plane having N.times.M buckets each bucket corresponding to one objective numeric attribute value, said plane having two axes respectively corresponding to two predictive numeric attributes of data included in said database and belonging to one node of said regression tree;
- storing the number of tuples of data in each bucket and the sum of said objective numeric attribute values of data in each bucket so as to correspond to each bucket in said plane;
- segmenting a bucket region of said plane that minimizes the mean squared error of said objective numeric attribute values from the plane;
- generating a first node for data inside of the segmented bucket region and a second node for data outside of the segmented bucket region;
- and
- producing said regression tree from said first and second nodes.
A method for generating a regression tree for an objective numeric attribute of data in a database, the method comprising:
- forming a plane having N.times.M buckets, each bucket corresponding to an objective numeric attributes value, said plane having two axes respectively corresponding to two predicative numeric attributes of data included in said database and belonging to one node of said regression tree;
- storing the number u(i,j) of data in each bucket (i,j) and the sum v(i,j) of said objective numeric attribute values of data in each bucket (i,j) so as to correspond to each bucket in said plane;
- generating a first node for data inside of the segmented bucket region and a second node for data outside of the segmented bucket region;
- and
- producing said tree from said first node and said second node.
An apparatus for generating a regression tree for an objective numeric attribute of data in a database, said apparatus comprising:
- a plane forming module for storing an N.times.M array of buckets, each bucket storing the number of tuples of data in each respective bucket and the sum of said objective numeric attribute values of data in each respective bucket, said plane having two axes respectively corresponding to two predicative numeric attributes of data included in said database and belonging to one node of said regression tree, each bucket corresponding to an objective numeric attribute value;
- a segmenting module for segmenting a bucket region that minimizes the mean-squared error of said objective numeric attribute values;
- a module for generating nodes for data inside of the segmented bucket region and data outside of the segmented bucket region;
- and
- a module for producing the regression tree from the generated nodes.
An apparatus for generating a regression tree for an objective numeric attribute of data in a database, said apparatus comprising:
- a plane forming module for storing an N.times.M array of buckets, each bucket storing the number u(i,j) of data in each respective bucket (i,j) and the sum v(i,j) of said objective numeric attribute values of data in each respective bucket (i,j), said plane having two axes respectively corresponding to two predicative numeric attributes of data included in said database and belonging to one node of said regression tree;
- a segmenting module for segmenting a bucket region that maximizes an interclass variance from the plane;
- a module for generating a first node for data inside of the segmented bucket region and a second node for data outside of the segmented bucket region;
- and
- a module for producing the regression tree from the first and second nodes.
A storage medium storing a program for generating a regression tree for an objective numeric attribute of data in a database, said program causing a computer to execute functions comprising:
- forming a plane having N.times.M buckets, each bucket corresponding to an objective numeric attribute value, said plane having two axes respectively corresponding to two predicative numeric attributes of data included in said database and belonging to one node of said regression tree;
- storing the number of data in each bucket and the sum of said objective numeric attribute values of data in each bucket so as to correspond to each bucket in a plane;
- segmenting a bucket region that minimizes the mean-squared error of values of said objective numeric attribute from the plane;
- generating a first node for data inside of the segmented bucket region and a second node for data outside of the segmented bucket region;
- and
- producing said regression tree from said first node and said second node.
A storage medium storing a program for generating a regression tree for an objective numeric attribute of data in a database, said program causing a computer to execute functions comprising:
- forming a plane having N.times.M buckets, each bucket corresponding to an objective numeric attribute value, said plane having two axes respectively corresponding to two predicative numeric attributes of data included in said database and belonging to one node of said regression tree;
- storing the number u(i,j) of data in each bucket (i,j) and the sum v(i,j) of said objective numeric attribute values of data in each bucket (i,j) so as to correspond to each bucket in said plane;
- segmenting a bucket region that maximizes an interclass variance from the plane;
- generating a first node for data inside of the segmented bucket region and a second node for data outside of the segmented bucket region;
- and
- producing said tree from said first node and said second node.
Dependencies are inferred from the claim text.