Summary

After introducing deep learning in the previous chapter, in this chapter, we gained a deeper understanding of the common architectural blocks that are used for time series forecasting. The encoder-decoder paradigm was explained as a fundamental way we can structure a deep neural network for forecasting. Then, we learned about FFNs, RNNS, and CNNs and explored how they are used to process time series. We also saw how we can use all these major blocks in PyTorch by using the associated notebook and got our hands dirty with some PyTorch code.

In the next chapter, we’ll learn about a few major patterns we can use to arrange these blocks to perform time series forecasting.