Welcome to Time Series Foundations, the first module of the course. Before any forecasting model, you need to understand why time series data is different — because the habits you built on ordinary tabular data will actively mislead you here. In a normal dataset, rows are independent and interchangeable; you can shuffle them, split them randomly, and each row stands alone. A time series is the opposite: the order is the information, each value leans on the ones before it, and shuffling destroys the very structure you want to model. Getting this foundation right is what keeps the rest of the course honest.

You’ll start with what makes time series special — temporal order, autocorrelation, and why standard cross-validation leaks the future. You’ll learn to handle a series in pandas with a datetime index, resampling it to different frequencies and smoothing it with rolling windows. You’ll see how to explore a series visually to spot trend and seasonality. And you’ll learn the one split that’s valid for forecasting — a chronological train/test split — plus the naive baselines every real model must beat. The module ends with a guided project meeting Cyclepath, the bike-share series you’ll forecast for the rest of the course.

Every example is real, runnable Python on the seeded Cyclepath series — the resampling, rolling averages, chronological split, and baseline forecasts all produce the exact numbers you’ll see, and you can reproduce them with pandas and numpy. Start with Lesson 1 on what makes time series special.