​
Guide to Automated Prediction Engineering

​
Introduction

Trane is a comprehensive machine learning library designed for the effortless generation of prediction problems and the subsequent creation of labels for supervised learning applications. Built with the intent to simplify and accelerate the data preparation phase, Trane aims to propel the automation frontier of the machine learning process.

In the world of data science, researchers and professionals often deal with numerous records of specific entities. The main objective is frequently to predict future occurrences or behaviors of these entities. Trane’s unique design focuses on these time-related prediction challenges, converting your metadata into actionable prediction problems and well-defined cutoff times.

​
Understanding Prediction Problems & Cutoff Times

At its core, a prediction problem in Trane is structured using a designated formal language, which we’ll delve into below. The concept of ‘cutoff times’ is pivotal in Trane’s functionality. It represents the final timestamp in your dataset that’s earmarked for training your model. Any subsequent data following this cutoff time is strictly reserved for evaluating the model’s performance, ensuring there’s no overlap between training and testing phases.

​
Key Features of Trane:

  1. Prediction Tasks Generation:
    • Trane seamlessly transforms column names and types into prediction tasks, presenting them in a human-readable format.
  2. Labeling:
    • Trane’s advanced system creates accurate labels based on the prediction tasks. These labels encompass historical occurrences of specific events.
    • These labels are primed and ready, making it easy for you to integrate them directly into various machine learning algorithms.

To encapsulate, Trane is your go-to tool for transforming raw, structured data into actionable prediction problems. By auto-generating the associated labels, Trane equips you with the right foundation to craft a robust feature matrix for your machine learning model, streamlining and enhancing your data science journey.