R Deep Learning Projects

Implementation without R6

In this section, we include the implementation of the same basic recurrent neural network without using R6 classes. First, some imports and setting the seed:

library(readr)library(stringr)library(purrr)library(tokenizers)set.seed(1234)

We introduce an auxiliary function to initialize to zeros a matrix with the shape of a matrix, M:

zeros_like <- function(M){ return(matrix(0,dim(as.matrix(M))[1],dim(as.matrix(M))[2])) }

We also need the softmax function:

softmax <- function(x){ xt <- exp(x-max(x)) return(xt/sum(xt))}

We will use this for testing the female names data (see the Exercises section):

data <- read_lines("./data/female.txt")

And do some preprocessing:

text <- data %>% str_to_lower() %>% str_c(collapse = ...