数据驱动的自动化机器学习流程生成方法

为有效解决自动化机器学习的数据处理时间较长等问题，文中提出了数据驱动的自动化机器学习流程生成方法。该方法的数据集匹配模块通过余弦相似性来计算数据集之间的相关性，获取和当前数据集时间匹配程度最高的数据集，并将其和历史数据搜索结果相结合，先选择最优选择动作，扩展和选择MCTS节点。该方法引入了服务关联关注约束模型，可以依据历史数据对所有的存在关联约束强度的选择动作进行指导，生成自动化的机器学习流程。测试结果显示，该方法对不同数据集的学习流程生成耗时均低于580s，生成的流程对单标签和多标签的数据分类结果均在0.92以上，数据完整度均高于95.5%，应用性良好。

Data-driven Automatic Machine Learning Process Generation Method

In order to effectively solve the problems of long data processing time in automated machine learning, a data-driven automated machine learning process generation method is proposed in this paper. The dataset matching module of this method calculates the correlation between datasets through cosine similarity, obtains the dataset with the highest degree of time matching with the current dataset, and combines it with the historical data search results. First select the optimal selection action, expand and select the MCTS node. The method introduces a service association attention constraint model, which can guide all selection actions with association constraint strength based on historical data, and generate an automated machine learning process. The test results show that the learning process generation time of this method for different datasets is lower than 580s, the generated process classifies the data of single label and multi-label data above 0.92, and the data completeness is higher than 95.5%. Good applicability, and the application is good.