PQL: A Declarative Language for Predictive Modeling on Relational Databases
The Predictive Query Language (PQL) is a new SQL-inspired declarative language designed to define predictive tasks on relational databases. It aims to automate the extraction of training examples—prediction entities and target labels—from databases, a process that is currently manual, slow, and error-prone. PQL allows users to specify a predictive task in a single declarative query, enabling automatic computation of training data for machine learning models. These models are used in applications such as recommendations, financial fraud detection, and supply chain optimization, providing billions of predictions daily. The language is presented in a paper on arXiv (2602.09572) and targets the inefficiencies of manual data preparation for predictive modeling.
Key facts
- PQL is an SQL-inspired declarative language for predictive modeling on relational databases.
- It automates the extraction of training examples (prediction entities and target labels) from databases.
- Predictive models are used in recommendations, fraud detection, and supply chain optimization.
- The language allows specifying a predictive task in a single declarative query.
- The paper is available on arXiv with ID 2602.09572.
- Current manual extraction is slow, laborious, and prone to mistakes.
- Predictive models provide billions of predictions daily.
- PQL enables automatic computation of training data for machine learning.
Entities
Institutions
- arXiv