RNNs from scratch in R
The purpose of this section is to show you how you can implement recurrent neural networks from bare bones in R. This is perhaps not the optimal solution for a number of reasons, but it is a great way to get started in deep learning.Â
There are many plug and play frameworks like H2O, MXNet, TensorFlow, or Keras, that have compatibility with R. Our goal is to focus on the understanding of the algorithm rather than a particular API, although we will include an example using Keras. This is for two reasons, at the time of writing, the compatibility with R suffers from growing pains and we encountered many errors and issues with the different packages. On the other hand, even the stable versions of such packages have ever-changing APIs. We will focus on this section in building a very simple recurrent neural network from scratch, using simple tools from R.
We will start from the beginning, with a super-quick introduction to R6 classes in R using the example of the perceptron...
