Feature processing method and system for machine learning

Abstract

One of the embodiments relates to a feature processing method and a system for machine learning. The method comprises the steps of obtaining a plurality of candidate features from a basic feature setand obtaining a plurality of candidate operators from a basic operator set, and forming a plurality of initial feature combinations by the plurality of candidate features and the plurality of candidate operators; taking the plurality of initial feature combinations as an initial population of a genetic programming algorithm, and performing genetic manipulation on the initial population by adoptingthe genetic programming algorithm to obtain an optimized target population; obtaining a target feature combination based on the optimized target population, wherein the target feature combination isrepresented by calculation results of basic features and basic operators; wherein the basic features belong to a basic feature set, and the basic operators belong to basic operator subsets; and takingthe target feature combination as a feature of machine learning to participate in operation of machine learning.

Description

technical field [0001] The embodiments of this specification relate to the field of artificial intelligence, and in particular to a feature processing method and system for machine learning. Background technique [0002] Features are the basis of all machine learning models, and good features play a vital role in the effect of the model. The expressive ability of a single feature is often limited, and combining multiple features can improve the expressive ability. Feature mining itself is a relatively cumbersome task. The feature combinations obtained through manual mining are often only 3-4 orders. The feature combinations obtained by other automatic mining feature combination methods are implicit feature combinations. Such feature combinations are consistent with the target. The relationship between variables is very hidden, and the interpretability is low, or the obtained feature combination is a semi-explicit feature combination. Although the relationship between such a...

AI Technical Summary

Problems solved by technology

Feature mining itself is a relatively cumbersome task. The feature combinations obtained through manual mining are often only 3-4 orders. The feature combinations obtained by other automatic mining feature combination methods are implicit feature combinations. Such feature combinations are consistent with the target. The relationship between variables is very hidden, and the interpretability is low, or the obtained feature combination is a semi-explicit feature combination. Although the relationship between such a feature combination and the target variable is not as hidden as the implicit feature combination, its limited expressiveness

Images