Chapter 13: GPU Accelerated Model for Multivariate Forecasting

Time series analysis models can become quite large, and their training can become computationally expensive. This is especially the case when moving from univariate to multivariate time series predictions. In some of these cases, GPUs can be used to accelerate the process.

So far, we have described univariate time series models, relying on past values of one time series to predict the future value of the same time series. Often, real-world problems and data are not that simple. If two time series correlate, the value of a variable at a given point in time can depend on the past values of the same variable and the past values of other variables from other series. Integrating such multiple variables into the input or output of the model is called multivariate time series analysis. A model for multivariate predictions can output just one value, which is the next value of one of the time series, or multiple values, which...