机器学习除了要关注传统意义上的学习方法和算法精度等外,还需要关注机器学习的易用性。针对易用性,提出了一种对外提供restful接口服务的机器学习web服务引擎。通过对机器学习算法、模型和优化器等进行封装,屏蔽掉复杂的参数选取优化过程,实现简化机器学习的运用的目的。最后以垃圾邮件分类应用和漫画推荐应用为例,结构化输入训练数据,通过查询得到预测结果,完成邮件预测分类和漫画推荐功能。实验结果表明框架能搭载不同机器学习模块实现不同应用,验证了服务引擎的功能,简易的实现机器学习服务。解决了机器学习易用性的问题。

In addition to focusing on the learning methods and the algorithm precision in the traditional sense, machine learning needs to pay attention to the ease of use. Aiming at the ease of use, this paper proposes a machine learning web service engine that provides restful interface services externally. By encapsulating machine learning algorithms, models and optimizers, the complex parameter selection and optimization process is shielded, simplifying the use of machine learning. Finally, taking the application of spam classification and comic recommendation as an example, the training data are input structurally, and the prediction results are obtained by query to complete the mail classification and comic recommendation functions. The experimental results show that the framework can carry different machine learning modules for different applications, verify the functions of the service engine and implement the machine learning service easily.