Vector data high-order feature optimal transformation method based on interaction detection

Abstract

The invention provides a vector data high-order feature optimal transformation method based on interaction detection. The method comprises the following steps: acquiring a teacher model; performing binary feature interaction detection on the teacher model to obtain binary feature interaction pairs; performing hierarchical high-order feature interaction pair detection according to the binary feature interaction pair to obtain a high-order feature interaction pair; constructing a parameterized neural network model according to the binary feature interaction pair and the high-order feature interaction pair; and solving the neural network model to obtain optimal parameters. A teacher model is obtained through training, a binary feature interaction pair and a high-order feature interaction pair are detected through feature interaction in a classroom model, then interaction information capable of capturing features is constructed, a neural network model is solved, optimal parameters are obtained, the goodness of the model is obviously improved, the optimal transformation of the features or the feature combination can be given in a parameterized form, and more accurate prediction can be conveniently obtained.

Description

Technical field [0001] The present application relates to the field of feature engineering, and in particular to a high-order feature optimal transform method based on interactive detection. Background technique [0002] In the field of machine learning, the research on optimal transformation is relatively rare, the most important reason is to design the difficulty of designing a valid algorithm. [0003] Traditional methods utilize the random combination of features and some basic calculations (plus-reduction-multiply-divided) and several primary transformation functions such as triangular functions, power functions, polynomial functions, etc.). With this new feature, even if the number is huge, it is difficult to "guess" a hidden (i.e. optimal) characteristic transformation, the traditional algorithm constructs a large number of useless new features, but in most cases, it is impossible to improve The role of subsequent learning task performance, but a lot of calculation resourc...

AI Technical Summary

Problems solved by technology

[0008] This application provides an optimal transformation method for high-order features of vector data based on interaction detection to solve the problem that the classic ACE algorithm will give poor fitting and prediction results when the features have high-dimensional interactions in the prior art The problem

Images