Lavoisier: High-Level Selection and Preparation of Data for Analysis

摘要

Most data mining algorithms require their input data to be provided in a very specific tabular format. Data scientists typically achieve this task by creating long and complex scripts, written in data management languages such as SQL, R or Pandas, where different low-level data transformation operations are performed. The process of writing these scripts can be really time-consuming and error-prone, which decreases data scientists' productivity. To overcome this limitation, we present Lavoisier, a declarative language for data extraction and formatting. This language provides a set of high-level constructs that allow data scientists to abstract from low-level data formatting operations. Consequently, data extraction scripts' size and complexity are reduced, contributing to an increase of the productivity with respect to using conventional data manipulation tools.