AutoML 2.0必将令数据科学更加自动化

时间：2022-03-25 11:09:01 | 来源：行业动态

时间：2022-03-25 11:09:01 来源：行业动态

First-generation AutoML platforms have focused on automating the machine learning part of the data science process. In a traditional data science workflow, however, the longest and most challenging part is the highly manual step known as feature engineering. Feature engineering involves connecting data sources and building a flat "feature table" with a rich, diverse set of "features" that is evaluated against multiple Machine Learning algorithms. The challenge of feature engineering is that it requires an elevated level of domain expertise to ideate new features and is very iterative as features are evaluated and rejected or chosen. New platforms, however, have recently emerged that provide additional capabilities and automation aimed at solving this challenge. Platforms with "Automated Feature Engineering" capabilities now allow for the automated creation of feature-tables from relational data sources as well as flat files. This ability to "auto-generate" features in the data science process is a game-changing capability. Suddenly, the "citizen" data scientists - Business Intelligence (BI) analysts, data engineers, and other technically savvy members of the organization with deep domain knowledge - can become valuable contributors to an organization's development of ML and AI models. Through Automated Feature Engineering, BI teams can suddenly develop sophisticated predictive analytics algorithms in days, significantly accelerating their productivity with minimal help from data scientists.

第一代AutoML平台的重点主要放在自动化数据科学过程中的机器学习部分。但在传统的数据科学工作流程里，最冗长和最具挑战性的部分则是被称之为是要素工程的部分，要素工程是高度手动的一步，主要涉及到连接数据源及构建宽大的要素表，需包含丰富多样的要素。与此同时，这些要素还需要针对多种机器学习算法进行评估。

目前，要素工程面临的挑战是，只有用更高水平领域的专业知识才能酝酿新的要素，而且这一过程需要在评估、拒绝或选择要素时反复地做。但最近业界出现了新平台，这些新平台可以提供旨在解决这一挑战的附加功能及自动化功能。现在一些具有自动要素工程功能的平台可以从关系数据源以及无结构文件里自动创建要素表。这种能够在数据科学过程中自动生成要素的方法，可以说是个改变游戏规则的功能。

于是，突然之间，公民数据科学家开始成为组织开发ML和AI模型的有价值贡献者。一般来说，「公民数据科学家」指的是商业智能（BI）分析师、数据工程师和组织中其他具有深厚领域知识的、精通技术的成员。借助于机器学习，BI团队利用自动化要素工程可以在几天之内开发出复杂的预测分析算法，无需数据科学家帮忙就可以极大地提高生产力。

**Automating Data Science: Democratization**