README.md

In germayneng/DogEncode: Easy categorical encoding for machine learning feature engineering

DogEncode

DogEncode provides some methods to encode categoical features fast and easy. From simple methodology like dummy variables to Owen Zhang's kaggle famous leave one out encoding, this R package will cover them all.

Installation

# install.packages("devtools")
devtools::install_github("germayneng/DogEncode")

Usage

Currently, DogEncode covers the following encoding methods. I will be glad to add more as I learn more :) 1) one hot encoding (dummy)

# feed in dataframe wtih cat variables. class of variables to be character 
# mode = auto to automate dropping off one dummy 
# mode = manual to generate all dummy 
one_hot_encoding(df, mode = "manual") # you can bind this entire dataframe to your features 

2) label encoding aka colhot encoding

# feed in dataframe with cat variable. class of variables to be character 
# if you have both train and test, you can feed both inside to ensure that the labels are standardized between them 
# otherwise, let argument test = NULL 

# example if there is test 
end <- label_encoding(train = temp, test = temp_test) 
Names <- end[[1]] # encoded for train 
Names_test <- end[[2]] # encoded for test 

# if there is no test 
end <- label_encoding(train = temp, test = NULL) 

3) Count encoding aka counthot encoding

# feed in single column of variable, i.e train$cat 
# example 

example <- data.frame(
   id = c(rep("a", 5), rep("b", 3), rep("c", 2), "d"),
   resp = c(1, 0, 0, NA, 0, 1, 0, 0, 1, 0, 1))


 count_encoding(example$id, mode = "normal")

 # if want log scale 
 count_encoding(example$id, mode = "log")

4) Label count encoding

example <- data.frame(
   id = c(rep("a", 5), rep("b", 3), rep("c", 2), "d"),
   resp = c(1, 0, 0, NA, 0, 1, 0, 0, 1, 0, 1))


label_count_encoding(example$id)

5) Leave one out encoding

example <- data.frame(
   id = c(rep("a", 5), rep("b", 3), rep("c", 2), "d"),
   resp = c(1, 0, 0, NA, 0, 1, 0, 0, 1, 0, 1))

 LOO_encode(example$id, example$resp) 

To do

• Add more documentation to explain every function and credit to respective people.
• Hash encoding

Credit

Owen Zhang, HJ van Veen, Dex Grooves and Simone Aiosa's dog logo super helpful post to help debug ^.^

License

germayneng/DogEncode documentation built on Sept. 2, 2019, 11:02 a.m.