Learn how to optimize model hyperparameters and even the architecture in a few lines of code
In my previous article, we explored the basics of time series forecasting with sktime, looking at how to leverage this powerful library for straightforward forecasting tasks. Now, it’s time to take our journey further and dive into the advanced techniques that can help you optimize your forecasts and improve their accuracy. In this follow-up, we’ll explore how to build more sophisticated models, tune hyperparameters, and even do model architecture search with sktime.
First, for an easy start, let me demonstrate the basic sktime workflow again. This time, we will use the Longley dataset, which is part of sktime (BSD-3 license). It contains various US macroeconomic variables from the years 1947 to 1962 and looks like this: