Methods and systems for relating features with labels in electronics

A system, comprising:

• anon-transitory computer-readable memory medium;
• and
• a processor coupled to the memory medium, wherein the processor is configured to:
• obtain a training set of data comprising a plurality of sets of data each representative of an electronic item, each set comprising feature values for a plurality of features, and for at least one label;
• build a reduced training set of data which comprises an aggregated representation of the training set of data, the building comprising:
• dividing the sets of data into a plurality of groups, wherein each set of data is assigned to one group, wherein all sets of data for which feature values meet at least one similarity criterion are in the same group, wherein the at least one similarity criterion comprises feature values for at least one feature not differing by more than a threshold amount between sets of data of the group;
• and
• storing in the reduced training set of data, for each group, an aggregated set of data comprising:
• an aggregated representation of feature values for the sets of data of the group, wherein the aggregated representation of feature values comprises an average of the feature values for the sets of data of the group,
• an aggregated representation of at least one label value of the sets of data of the group, wherein the aggregated representation of the at least one label value comprises a sum of label values for the sets of data of the group, and
• a number of sets of data of the respective group, wherein the number of sets of data is less than the total number of sets of data of the respective group;
• and
• randomize the number of sets of data of each of the plurality of groups of the reduced training set of data a plurality of times in order to obtain a plurality of randomized reduced training sets of data, wherein randomizing the number of sets of data is performed based on a respective probability distribution characterizing a respective probability to obtain different numbers of sets of data of the respective group;
• use the plurality of randomized reduced training sets of data in a classification algorithm implementing a plurality of decision trees for determining a relationship between the at least one label and the features of the electronic items, wherein each randomized reduced training set of data is used in a different decision tree of the plurality of decision trees of the classification algorithm for determining the relationship between the at least one label and the features of the electronic items based on respective outputs of the plurality of decision trees;
• and
• store an output of the classification algorithm in the non-transitory computer-readable memory medium.

A non-transitory storage device readable by a machine, tangibly embodying a program of instructions executable by the machine to perform operations comprising:

• obtaining a training set of data comprising a plurality of sets of data each representative of an electronic item, each set comprising feature values for a plurality of features, and for at least one label;
• building a reduced training set of data which comprises an aggregated representation of the training set of data, the building comprising:
• dividing the sets of data into a plurality of groups, wherein each set of data is assigned to one group, wherein all sets of data, for which feature values meet at least one similarity criterion, are in the same group, wherein the at least one similarity criterion comprises feature values for at least one feature not differing by more than a threshold amount between sets of data of the group,
• storing in the reduced training set of data, for each group, an aggregated set of data comprising:
• an aggregated representation of feature values for the sets of data of the group, wherein the aggregated representation of feature values comprises an average of the feature values for the sets of data of the group,
• an aggregated representation of at least one label value of the sets of data of the group, wherein the aggregated representation of the at least one label value comprises a sum of label values for the sets of data of the group, and
• a number of sets of data of the respective group, wherein the number of sets of data is less than the total number of sets of data of the respective group;
• and
• randomizing the number of sets of data of each of the plurality of groups of the reduced training set of data a plurality of times in order to obtain a plurality of randomized reduced training sets of data, wherein randomizing the number of sets of data is performed based on a respective probability distribution characterizing a respective probability to obtain different numbers of sets of data of the respective group;
• using the plurality of randomized reduced training sets of data in a classification algorithm implementing a plurality of decision trees for determining a relationship between the at least one label and the features of the electronic items, wherein each randomized reduced training set of data is used in a different decision tree of the plurality of decision trees of the classification algorithm for determining the relationship between the at least one label and the features of the electronic items based on respective outputs of the plurality of decision trees;
• and
• storing an output of the classification algorithm in the non-transitory computer-readable memory medium.