How to one hot encode several categorical variables in R

Question

I\'m working on a prediction problem and I\'m building a decision tree in R, I have several categorical variables and I\'d like to one-hot encode them consistently in my tra

Answer 1

I have a tidy solution that gives more control to user over the entire process. My solution has a JavaScript component that splits each cell and stores the column names as JSON. Then I use tidyjson::spread_all function to spread JSON into different column names.

JavaScript component that you need to save as encoder.js:

function oneHotSplitEncoder(inputStrArray, prefix, spliterRegExStr, spliterRegExStrOptions){
  if (Array.isArray(inputStrArray)) {
    return inputStrArray.map(function(str) {
      try{
        if(typeof(str) === 'string' && typeof(spliterRegExStr)==='string' && typeof(spliterRegExStrOptions)==='string' && typeof(prefix) === 'string'){
          return JSON.stringify(
            str.split(RegExp(spliterRegExStr, spliterRegExStrOptions))
               .reduce(function(p, component){
                 p[prefix + component] = 1;
                   return p;
               }, {})
          )
        } else {
          return NaN;
        }
      } catch (e) {
        console.warn("\n"+e+"\n"+str+"\n"+spliterRegExStr+' string expected')
        return NaN;
      }
    });
  } else {    
    console.warn("Error: oneHotSplitEncoder function needs array type inputs");
    return NaN;
  }
};

R components:

library('dplyr')
js <<- V8::v8(); 
js$source("encoder.js");
oneHotSplitEncoder <- function(inputStrArray, prefix, spliterRegExStr, spliterRegExStrOptions)
  js$call("oneHotSplitEncoder", inputStrArray, prefix, spliterRegExStr, spliterRegExStrOptions)

df_one_hot <- df %>%
  mutate(
    fooColumn = oneHotSplitEncoder(fooColumn, 'prefix.', ' *[,;] *', 'g')
  ) %>%
  bind_cols(tidyjson::spread_all(.$fooColumn) %>% select(-document.id) %>% replace(is.na(.), 0))