Machine learning classification method and device

Abstract

The invention discloses a machine learning classification method and device. The machine learning classification method comprises the steps that training is conducted through training configuration files and training samples, and at least one classifier model is obtained; the training configuration files comprise data format definition, at least one classification task and task parameters of each classification task, and the training samples comprise at least one default property according with the data format definition; classification configuration files and data to be classified are received, wherein the classification configuration files comprise a classification topological structure, at least one classifier model trained in advance and the data format definition, and the data to be classified comprise at least one property according with the data format definition; a multi-stage classification structure is built according to the classification topological structure and the classifier model; the classes of the data to be classified are judged step by step through the multi-stage classification structure. Therefore, the multi-stage classification structure is easily and conveniently built, and no code development exists in the process.

Description

technical field [0001] The invention relates to the field of network technology, in particular to a machine learning classification method and device. Background technique [0002] Machine learning classification is a multi-field interdisciplinary subject that can be applied in text mining, machine translation, artificial intelligence and search engines. Currently, widely used classifiers include Libsvm, decision tree C5.0, and naive Bayesian. [0003] At present, each classifier is used independently and does not meet the conditions for direct compatibility with other classifiers. For example, if C5.0 and Libsvm are used to construct a two-level classifier, additional splicing is required. The form of machine learning classification is relatively simple, and there is a lack of complex forms for connecting, comparing and comprehensively using various classifiers, which restricts the classification effect. [0004] In addition, different classification algorithms have differ...

AI Technical Summary

Problems solved by technology

[0003] At present, each classifier is used independently and does not meet the conditions for direct compatibility and combination with other classifiers. For example, if C5.0 and Libsvm are used to construct a two-level classifier, additional splicing is required. The form of machine learning classification is relatively simple, and there is a lack of complex forms for connecting, comparing and comprehensively using various classifiers, which restricts the classification effect

[0004] In addition, different classification algorithms have different ways of understanding data, and developers have different design tendencies and interface encapsulations. As a result, most classifiers have different data input and output formats, and their compatibility is poor.

In order to realize the mutual compatibility of the classifiers, format converters need to be produced in pairs, and the switching cost is high

Images