f_model_plot_variable_dependency_regression: plot model dependency on most important variables

In erblast/oetteR: Collection of personal R functions

Description

response variable will be plotted against the entire range of each variable staring with the most important ones. All other variables will be set to median or most common factor. This function requires a ranked list of the most important variables as returned by f_model_importance()

Usage

f_model_plot_variable_dependency_regression(m, ranked_variables,
  title = unlist(stringr::str_split(class(m)[1], "\\."))[1],
  data = NULL, formula, data_ls,
  variable_color_code = f_plot_color_code_variables(data_ls),
  limit = 12, log_y = F, set_manual = list(), ...)

Arguments

m	a regression model
ranked_variables	datafram as returned by f_model_importance()
title	character vector as plot title, Default: unlist(stringr::str_split(class(m)[1], "\."))[1]
data	a dataframe, only necessary if it differs from data_ls$data, Default: NULL
formula	the formula used to train the model
data_ls	data_ls object generated by f_clean_data(), or a named list list( data = <dataframe>, numericals = < vector with column names of numerical columns>)
variable_color_code	dataframe created by f_plot_color_code_variables()
limit	integer limit the number of variables to be plotted, Default: 12
log_y	boolean log_scale for y axis
set_manual	named list, set some variables manually instead of defaulting to median or most common factor. !! Values need to be of the same variable type as in the original data.
...	arguments passed to facet_wrap e.g. usefull for nrow, ncol

Value

plot

See Also

str_split

Examples

# regular version--------------------------------------
data_ls             = f_clean_data(mtcars)
data                = data_ls$data
formula             = disp~hp+mpg+cyl
m                   = randomForest::randomForest(formula, data)
ranked_variables    = f_model_importance( m, data)
variable_color_code = f_plot_color_code_variables(data_ls)
limit               = 12
f_model_plot_variable_dependency_regression( m
                                             , ranked_variables
                                             , title = unlist( stringr::str_split( class(m)[1], '\\.') )[1]
                                             , formula = formula
                                             , data_ls = data_ls
                                             , variable_color_code = variable_color_code
                                             , limit = limit
                                             )

#pipe version ------------------------------------------

data_ls = f_clean_data(mtcars)
form = as.formula('disp~hp+cyl+wt')
variable_color_code = f_plot_color_code_variables(data_ls)
limit            = 10

 pl = pipelearner::pipelearner( data_ls$data ) %>%
  pipelearner::learn_models( rpart::rpart, form ) %>%
  pipelearner::learn_models( randomForest::randomForest, form ) %>%
  pipelearner::learn_models( e1071::svm, form ) %>%
  pipelearner::learn() %>%
  mutate( imp   = map2(fit, train, f_model_importance)
          ,plot = pmap( list( m = fit, ranked_variables = imp, title = model, data = train)
                        , .f = f_model_plot_variable_dependency_regression
                        , formula = form
                        , data_ls = data_ls
                        , variable_color_code = variable_color_code
                       , limit = limit
         )
  )

erblast/oetteR documentation built on May 27, 2019, 12:11 p.m.