Customisation

The CITRUS package has been built in a modular way, so that new approaches can be added easily. There are 2 ways to run a new approach, within segments or by creating a new approach for a current module that follows the same structure as the module that you are looking to add/change.

To begin, customising within segments has been made possible by the addition of function arguments in the function calls for the modules. Currently, there is the option to add a custom function for the model step and the preprocessing step only, future development could involve addition of new custom functionality.

Below is an example of a custom function being added to modelling layer of the pipeline. This uses a user defined function to classify the customers, it skips the ‘preprocess’ step and goes straight from the ‘model’ step and onwards.

library(citrus)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

my_custom_function <- function(df) {
  
  final_df <- df %>%
    mutate(first_letter = substring(top_country, 1, 1)) %>%
    mutate(segment = ifelse(match(tolower(first_letter), tolower(LETTERS)) <= 8, 'Segment A',
                                      ifelse(match(tolower(first_letter), tolower(LETTERS)) <= 17, 'Segment B', 'Segment C'))) %>%
    arrange(desc(id))
  
  #output <- list('predicted_values' = final_df[,c('segment')])
  output <- list('predicted_values' = final_df[,c('id', 'segment')])
  
  return(output)
}
output <- segment(citrus::preprocessed_data, 
                  modeltype = 'tree',
                  FUN = my_custom_function,
                  FUN_preprocess = NULL,
                  steps = c('model'))                  

There is scope for new approaches to be created within the pre-processing, modelling and output layer steps. For example, a DBScan approach could be added to the modelling section as an alternative segmentation approach.