Time Series Forecasting in production

The seminar will show how data scientists work in real-life scenarios and the process they set up to ensure rapid iteration on a problem. Given a problem statement of a power-load forecasting problem, attendants will learn how to approach a machine learning problem, from the definition of the error metrics and baselines to how to structure your code to deploy a maintainable and easy-to-evolve solution. The goal is to showcase the steps to move from the experimentation phase to a production-grade solution, through the use of simple and well-established coding principles. In particular, the seminar will focus on:

1. A short introduction on what is a time series
2. Simple exploratory data analysis for time-series data
3. Understanding a problem statement and defining error metrics
4. Delivery of a hackable, working solution with a baseline model
5. Implementing a state-of-the-art model from the literature
6. Hands-on implementation and execution of a new model directly in the codebase

Speakers
Luca BAGGI, AI Engineer @xtream
Marco SIMONTE, AI Scientist @xtream

The seminar is part of the 2025/2026 edition of the Programming Laboratory course and is aimed at students in the second/third year of Undergraduate Degree Courses and the second year of Graduate Degree Courses, in particular for Degree Classes L-35 (Mathematical Sciences) and LM-91 (Techniques and Methods for the Information Society).

FOR INFORMATION

Prof. Daniele Toti
E-mail: daniele.toti@unicatt.it