Summary

In this chapter, we looked at the considerations of working with sequences. There are several tasks that require us to exploit information contained in a sequence, where sequence-agnostic models would fare poorly. We saw that using RNNs is a very powerful approach to sequence modeling – the architecture explicitly processes the sequence and considers the information accumulated so far, along with the new input, to generate the output. Even very simple RNN architectures performed very well on our stock price prediction task. We got the kind of results that would take a lot of effort to get using classical approaches.

We also saw that 1D convolutions can be employed in sequence prediction tasks. 1D convolutions, like their 2D counterparts for images, learn local features in a sequence. We built a 1D convolution model that didn't fare too well on our task. The final model that we built combined 1D convolutions and RNNs and provided excellent results regarding the...