Methods and systems of classifying multiple properties simultaneously using a decision tree

A computerized method of forming a decision tree to simultaneously classify multiple properties of a training item set, said training item set containing training items each having a plurality of X descriptors and a plurality of original properties, each of said plurality of X descriptors and said plurality of original properties corresponds to a physical aspect of a training item, said method comprising:

• defining X descriptor and property data for a plurality of training items, wherein at least some of said training items have a known value for a first property, and at least some of said items have a known value for a second property, wherein said second property is different from said first property;
• converting each of said training items into one or more converted training items, said converted training items each having said plurality of X descriptors, a K descriptor and a property;
• producing a pure-specific tree using a Bonferoni-modified t-test statistic splitting method, said pure-specific tree having a root node that splits said converted training items by a logic test on said K descriptor;
• growing a maximally generic tree, said maximally generic tree having a root node that splits said converted training items by a logic test on a X descriptor;
• and permuting said maximally generic tree to produce said decision tree, said decision tree operative to classify multiple unknown properties of a physical item.

A computerized method of using a decision tree to simultaneously predict multiple unknown properties of a physical item, said physical item having a plurality of X descriptors, said decision tree having a top layer of generic nodes that each splits by a logic test on one of said plurality of X descriptors, a middle layer of K nodes that each splits by a logic test on K, K representing property types of said multiple properties of said physical item, and a bottom layer of specific nodes that each splits by a logic test on one of said plurality of X descriptors, said method comprising:

• starting from a root node of said decision tree, evaluating said physical item against logic tests of said top layer of generic nodes and following a tree path resulted from said evaluating;
• evaluating said physical item against logic tests of said middle layer of K nodes, wherein a split produced by one of said middle layer of K nodes will differ depending on which of the multiple unknown properties is being predicted after the split such that multiple sub-paths resulting from said evaluating will be followed, each sub-path determined by which of the multiple properties is predicted at the end of said sub-path, each of said multiple sub-paths corresponding to each of said multiple unknown properties of said physical item;
• from each of said sub-paths, continue evaluating said physical item until each of said continued sub-paths reaches a respective terminal node;
• and
• for each of said unknown properties of said physical item, identify a value associated with each of said respective terminal nodes as a predicted value of said unknown property.

A computerized method of using a decision tree to predict multiple unknown properties of a physical item, said physical item having a plurality of X descriptors, said decision tree having a top layer of generic nodes that each splits by a logic test on one of said plurality of X descriptors, a middle layer of K nodes that each splits by a logic test on K, K representing property types of said multiple properties of said physical item, and a bottom layer of specific nodes that each splits by a logic test on one of said plurality of X descriptors, said method comprising:

• converting said physical item into one or more converted items, said converted items each having said plurality of X descriptors, a K descriptor having a value dependent on what unknown property of multiple unknown properties is being predicted by each converted item, and an unknown property, said unknown property being one of said multiple unknown properties, said converted items each having a different K descriptor value corresponding to a different unknown property;
• for each of said converted items, evaluating said converted item against said decision tree until a respective terminal node is reached;
• and
• for each of said unknown properties of said physical item, identify a value associated with each of said respective terminal nodes as a predicted value of said unknown property.

A computer system configured for forming a decision tree to simultaneously classify multiple properties of a training item set, said training item set containing training items each having a plurality of X descriptors and a plurality of original properties, each of said plurality of X descriptors and said plurality of original properties corresponds to a physical aspect of a training item, said system comprising:

• a converting module configured to convert each of said training item into one or more converted training items, said converted training items each having said plurality of X descriptors, a K descriptor having a value dependent on what unknown property of multiple unknown properties is being predicted by each converted item, and a property;
• an initial tree producing module configured to produce a pure-specific tree, said pure-specific tree having a root node that splits said converted training items by a logic test on said K descriptor;
• a tree growing module configured to grow a maximally generic tree, said maximally generic tree having a root node that splits said converted training items by a logic test on a X descriptor;
• and
• a tree permuting module configured to permute said maximally generic tree to produce said decision tree, said decision tree operative to classify multiple unknown properties of a physical item.

A computer system configured for using a decision tree to simultaneously predict multiple unknown properties of a physical item, said physical item having a plurality of X descriptors, said decision tree having a top layer of generic nodes that each splits by a logic test on one of said plurality of X descriptors, a middle layer of K nodes that each splits by a logic test on K, K representing property types of said multiple properties of said physical item, and a bottom layer of specific nodes that each splits by a logic test on one of said plurality of X descriptors, said system comprising:

• an evaluation module configured to:
• starting from a root node of said decision tree, evaluating said physical item against logic tests of said top layer of generic nodes and following a tree path resulted from said evaluating;
• evaluating said physical item against logic tests of said middle layer of K nodes, wherein a split produced by one of said middle layer of K nodes will differ depending on which of the multiple unknown properties is being predicted after the split such that multiple sub-paths resulting from said evaluating will be followed, each sub-path determined by which of the multiple properties is predicted at the end of said sub-path, each of said multiple sub-paths corresponding to each of said multiple unknown properties of said physical item;
• from each of said sub-paths, continue evaluating said physical item until
• each of said continued sub-paths reaches a respective terminal node;
• and
• an identification module configured to identify a value associated with each of said respective terminal nodes as a predicted value of said unknown property for each of said unknown properties of said physical item.