It’s About Time: Time-Series Forecasting with Darts

Track: Data Science

Type: Talk

Room: Lefthand Room (Seminar Room 3)

Time: Oct 03 (Thu): 10:25

Duration: 0:45

Along with the rise of “AI”, data-driven decision-making in businesses and organisations is also making a ‘buzz phrase’ of itself. And for most organisations, time series forecasting is a crucial component of making informed decisions. This talk will cover Darts, a relatively new Python library aimed at making time series analysis and forecasting more accessible while still including a range of models, from classical statistical methods to newer deep learning techniques, all without needing to import many other Python libraries.

In this talk, we will explore the tools Darts offers across the whole time series forecasting pipeline: preprocessing and anomaly detection, modelling (including ensembling), and performance evaluation and backtesting. We will start with an overview of the basic classes and components of Darts, followed by a demonstration of how to implement a basic ensemble model and visualise forecast results.

Additionally, we will also discuss some key features of Darts that sets it apart from more traditional time series libraries. In particular, the ‘Pythonic’ structuring of its models to use the fit() and predict() methods we’ve all come to know and love, similar to the popular Python scikit-learn library. We will also discuss some other advantages, such as the ease of including static past and future covariates, and encoders, as well as its integration with other popular Python libraries like pandas. As it’s a relatively new library, we will also discuss some of its current shortcomings.

Whether you’re a data scientist, analyst or simply interested in time series forecasting in Python (who isn’t?!), this talk will provide you with a solid foundational understanding of Darts so that you can apply it to your own work and projects.

Darts URL: https://unit8co.github.io/darts/