https://datatweets.com/courses/time-series-forecasting/time-series-foundations/Recent content in Time Series Foundations on DATATWEETSHugoenCopyright (c) 2025 DatatweetsSun, 28 Jun 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-1-what-makes-time-series-special/Fri, 10 Apr 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-1-what-makes-time-series-special/Ordinary datasets have independent, shuffleable rows. A time series is the opposite: temporal order carries meaning, values are autocorrelated, and standard cross-validation leaks the future into the past. This lesson explains the three properties that make forecasting its own discipline, and meets the Cyclepath series you’ll model all course.https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-2-the-datetime-index-and-resampling/Fri, 10 Apr 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-2-the-datetime-index-and-resampling/In pandas, a time series is a Series or DataFrame indexed by a DatetimeIndex that carries a frequency, and that calendar awareness is what makes real time-series work practical. This lesson builds the Cyclepath series, inspects its index, aggregates monthly data to yearly totals with resampling, and smooths it with a 12-month rolling mean that cancels the seasonality and reveals the trend.https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-3-exploring-and-visualizing-a-series/Fri, 10 Apr 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-3-exploring-and-visualizing-a-series/Summary statistics orient you, but they can’t show shape — two series with the same mean can look nothing alike over time. This lesson builds the first habit of forecasting: always plot the series first. You’ll read trend and seasonality off Cyclepath’s line plot, overlay a rolling mean to isolate the trend, and use a by-month view to confirm the seasonal period.https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-4-splitting-time-series-and-baselines/Fri, 10 Apr 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-4-splitting-time-series-and-baselines/Forecasting demands a chronological split — train on the past, test on the future — not a random one. This lesson holds out the last twelve months of Cyclepath, builds two baselines (naive and seasonal-naive), and scores them with MAE, RMSE, and MAPE, establishing the seasonal-naive bar that any real model must clear to earn its complexity.https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-5-guided-project-meet-the-cyclepath-series/Fri, 10 Apr 2026 09:00:00 +0200https://datatweets.com/courses/time-series-forecasting/time-series-foundations/lesson-5-guided-project-meet-the-cyclepath-series/The Module 1 capstone. You’ll build Cyclepath from code, explore its trend and yearly seasonality with summary stats, resampling, and a rolling mean, then hold out the last year and establish naive and seasonal-naive baselines. By the end you have the series, an honest test set, and a bar to beat: seasonal-naive at 5.9% MAPE.