Time series graphics

Line plots are the first visual aid in time series data analysis. T

We often have datasets with too many time series to be plotted together

In this case, we can rely on quantile plots or box-plots

A scatter plot can be used to visually assess the dependence of the target $y_t$ from exogenous inputs/features $x_{i, t}$.

Time embedding is helpful to see whether one can reduce the variance of segments in the time series to be predicted.

What is wrong with these time series of power measurements?

Since we are usually interested in more than one step ahead forecasts, we need good ways to visualize our model performances in more than one dimension.

Given a loss $l(y_{t+h}, \hat{y}_{t, t+h})$, where $h$ is the number of steps ahead (forecasting horizon), and $\hat{y}_{t, t+h}$ stands for the prediction done at time t for the step t+h, we can decide to investigate the loss as a function of h.