US 2019/0164060 A1Published application

Method of and server for converting a categorical feature value into a numeric representation thereof

AssigneeYandex Europe AG

InventorAndrey Vladimirovich Gulin

Life of this patent

Estimating the term…

Loading patent…

Abstract

There is disclosed a method of and a system for converting a categorical feature value into a numeric representation thereof, the categorical feature being associated with a training object used for training a Machine Learning Algorithm (MLA). The MLA is trained using several models, each model comprising a plurality of decision trees (an ensemble of decision trees). For each model, a respective set of training objects into an ordered list of training objects such that for each given training object there is at least one of: (i) a preceding training object that occurs before the given training object and (ii) a subsequent training object that occurs after the given training object. The method further comprises, for a given categorical feature, using a respective ordered set to generate numeric representation of values of the given categorical feature.

Claims20

Dependent claims

A method of converting a categorical feature value into a numeric representation thereof, the categorical feature being associated with a training object used for training a Machine Learning Algorithm (MLA), the MLA executable by a machine learning system to predict a target value for an in-use object, the method comprising:

accessing, from a non-transitory computer-readable medium of the machine learning system, a set of training objects, each training object of the set of training object containing a document and an event indicator associated with the document, each document being associated with a categorical feature;
generating a set of models for the MLA, each model of the set of models being based on an ensemble of decision trees;
for each model of the set of models:
organizing the set of training objects into a respective ordered list of training objects, the respective ordered list of training objects being organized such that for each given training object in the respective ordered list of training objects there is at least one of:
(i) a preceding training object that occurs before the given training object and
(ii) a subsequent training object that occurs after the given training object;
when building a given iteration of a decision tree in a given ensemble of decision trees:
selecting one of the set of models and the respective ordered list;
generating a decision tree structure using the one of the set of models;
when processing a given categorical feature using the decision tree structure, the given categorical feature associated with a given training object, the given training object having at least one preceding training object in the ordered list of training objects, generating a numeric representation thereof, the generating based on:
(i) a number of total occurrences of the at least one preceding training object with a same categorical feature value in the respective ordered list;
and
(ii) a number of pre-determined outcomes of events associated with at least one preceding training object having the same categorical feature value in the respective ordered list.

10 more elements

A method of converting a categorical feature value into numeric representation thereof, the categorical feature being associated with a training object used for training a Machine Learning Algorithm (MLA), the MLA executable by an electronic device to predict a value for an in-use object, the method comprising:

accessing, from a non-transitory computer-readable medium of the machine learning system, a set of training objects, each training object of the set of training object containing a document and an event indicator associated with the document, each document being associated with a categorical feature;
generating a set of models for the MLA, each model of the set of models being based on an ensemble of decision trees;
for each model of the set of models:
organizing the set of training objects into a respective ordered list of training objects, the respective ordered list of training objects being organized such that for each given training object in the respective ordered list of training objects there is at least one of:
(i) a preceding training object that occurs before the given training object and
(ii) a subsequent training object that occurs after the given training object;
when building a given iteration of a decision tree in a given ensemble of decision trees:
selecting one of the set of models and the respective ordered list;
generating a decision tree structure using the one of the set of models;
when processing a given categorical feature using the decision tree structure, for a given categorical feature, the given categorical feature associated with a given training object, the given training object having at least one preceding training object in the respective ordered list of training objects, generating a numeric representation thereof, the generating comprising calculating a function using a formula:
f(Number_WINs_PAST,Number_Occurence_PAST)
where,
Number_WINs_PAST comprises a number of pre-determined outcomes of events associated with at least one preceding training object having the same categorical feature values in the respective ordered set;
and
Number_Occurence_PAST comprises a number of total occurrences of the at least one preceding training object with a same categorical feature value in the respective ordered set.

12 more elements

A server configured to execute a Machine Learning Algorithm (MLA), the MLA being based on a decision tree prediction model based on a decision tree, the decision tree for processing a categorical feature value by converting it into a numeric representation thereof, the categorical feature being associated with a training object used for training a Machine Learning Algorithm (MLA), the MLA executable by the server to predict a target value for an in-use object, the server comprising:

a non-transitory computer-readable medium;
a processor coupled to the non-transitory computer-readable medium, the processor configured to:
access, from the non-transitory computer-readable medium of the machine learning system, a set of training objects, each training object of the set of training object containing a document and an event indicator associated with the document, each document being associated with a categorical feature;
generate a set of models for the MLA, each model of the set of models being based on an ensemble of decision trees;
to generate, the processor being further configured to, for each model of the set of models:
organize the set of training objects into a respective ordered list of training objects, the respective ordered list of training objects being organized such that for each given training object in the respective ordered list of training objects there is at least one of:
(i) a preceding training object that occurs before the given training object and
(ii) a subsequent training object that occurs after the given training object;
when building a given iteration of a decision tree in a given ensemble of decision trees, the processor being further configure to:
select one of the set of models and the respective ordered list;
generate a decision tree structure using the one of the set of models;
when processing a given categorical feature using the decision tree structure, for a given categorical feature, the given categorical feature associated with a given training object, the given training object having at least one preceding training object in the ordered list of training objects, generate a numeric representation thereof, the generating based on:
(i) a number of total occurrences of the at least one preceding training object with a same categorical feature value;
and
(ii) a number of pre-determined outcomes of events associated with at least one preceding training object having the same categorical feature value.

12 more elements

Dependencies are inferred from the claim text.