Machine learning hyper-parameter importance assessment method and system, and storage medium

Abstract

The present invention discloses a machine learning hyper-parameter importance assessment method and system, and a storage medium. The method comprises the steps of: obtaining different data sets in anOpenML, extracting meta features to show each data set, and collecting data of performances of a classification algorithm to be assessed in different hyper-parameter configurations; extracting meta features to show a used target data set, calculating distances between the meta features to obtain an increasing sequence of the distance between the target data set and a historical data set; using the data of the performances of the classification algorithm to be assessed in different hyper-parameter configurations to assess the hyper-parameter importance, executing a provided Relied and a clustering algorithm for former m historical data sets in order being close to the target data set according to an ordered sequence of distance increasing between the historical data set and the target dataset, and finally obtaining the hyper-parameter importance sequence of the classification algorithm to be assessed and a guided automatic parameter regulation process. The machine learning hyper-parameter importance assessment method and system give a certain guide for the hyper-parameter regulation of a classification algorithm black box so as to save the time and improve the purpose of efficiency.

Description

technical field [0001] The invention is a machine learning hyperparameter importance evaluation method, system and storage medium. Background technique [0002] Machine learning provides important technical support for data processing and data classification. However, model selection and parameter tuning are still two major problems that plague users, so automated machine learning systems emerged as the times require. The automated machine learning system uses automated machine learning algorithms to achieve automated data preprocessing, automated algorithm selection, and automated parameter adjustment, improving the accuracy of data classification and prediction, and freeing users from the heavy tasks of selecting algorithms and repeatedly adjusting parameters come out. [0003] Since the core of automated machine learning is automated algorithm selection and automated hyperparameter configuration, the system reduces the machine learning process to a Combined Algorithm Sel...

AI Technical Summary

Problems solved by technology

[0004] However, the configuration process of the hyperparameter configuration module of the current automated machine learning system is entirely based on experience, or the configuration of several hyperparameters is adjusted one by one by obtaining the final result through repeated iterations. Time, and repeated iterations also waste computer resources, and adjusting the configuration of all hyperparameters regardless of importance will waste user time and effort

Images