Welcome to this course on sequences and prediction, a part of the TensorFlow in practice specialization. In this course, we'll focus on time series, where you'll learn about different types of time series before we go deeper into using time series data. This week, you'll focus on time series themselves. We'll go through some examples of different types of time series, as well as looking at basic forecasting around them. You'll also start preparing time series data for machine learning algorithms. For example, how do you split time series data into training, validation, and testing sets? We'll explore some best practices and tools around that to get you ready for week 2, where you'll start looking at forecasting using a dense model, and how it differs from more naive predictions based on simple numerical analysis of the data. In week 3, we'll get into using recurrent neural networks to forecast time series. We'll see the stateless and stateful approaches, training on windows of data, and you'll also get hands-on in forecasting for yourself. Finally, in week 4, you'll add convolutions to the mix and put everything you've worked on together to start forecasting some real world data, and that's measurements of sunspot activity over the last 250 years. So let's get started with a look at time series, what they are, and the different types of them that you may encounter. Time series are everywhere. You may have seen them in stock prices, weather forecasts, historical trends, such as Moore's law. Here, I've plotted the number of transistors per square millimeter, where I grouped chip releases by year, and then drew the Moore's law trend line from the first data item, where you can see a correlation. If you want some really fun correlations, here's one from Tyler Vigen's site of Spurious Correlations. This one is a time series correlation of total revenue generated by video game arcades versus computer science doctorates awarded in the United States. While all of these are quite familiar, they begged the question, what exactly is a time series? It's typically defined as an ordered sequence of values that are usually equally spaced over time. So for example, every year in my Moore's law charts or every day in the weather forecast. In each of these examples, there is a single value at each time step, and as a results, the term univariate is used to describe them. You may also encounter time series that have multiple values at each time step. As you might expect, they're called Multivariate Time Series. Multivariate Time Series charts can be useful ways of understanding the impact of related data. For example, consider this chart of births versus deaths in Japan from 1950 to 2008. It clearly shows the two converging, but then deaths begin to outstrip births leading to a population decline. Now, while they could be treated as two separate univariate time series, the real value of the data becomes apparent when we show them together as a multivariate. Also consider this chart showing the global temperature against CO2 concentration. As univariates, they would show a trend, but when combined, the correlation is very easy to see adding further value to the data. Movement of a body can also be plotted as a series of univariates or as a combined multivariate. Consider, for example, the path of a car as it travels. A time step zero is at a particular latitude and longitude. As subsequent time steps, these values changed based on the path of the car. The acceleration of the car, in other words, it's not moving at a constant speed, means that the spaces between the time steps also change in size, in this case getting larger. But what if we were to plot the direction of the car as univariates. Based on its heading, we could see that the longitude of the car decreases over time, but its latitude increases, and as such you will get charts like these.