f_model_plot_var_dep_over_spec_var_range: plot vmodel varaible dependency over the range of a specified...

In erblast/oetteR: Collection of personal R functions

Description

Some models are able to capture relative dependencies. In order to visualise them the dataset is split into three parts. 0-25,25-75,75-100 percentile or the three most common factors.Then variable dependencies for each of the three splits are plotted. In the mtcars example below we can see that the model predicts an increase in disp if drat increases for cars with 8 cylinders, while the opposite is true for cars with only 6 cylinders.

Usage

f_model_plot_var_dep_over_spec_var_range(m, title, variables,
  range_variable, data, formula, data_ls, variable_color_code, log_y = F,
  limit = 12)

Arguments

m	a model
title	model title
variables	character vector with variable names, or ranked variables as returned by f_model_importance()
range_variable	character vector denoting range variable
data	dataset
formula	formula
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>) - The data_ls object provides the entire dataset
variable_color_code	dataframe created by f_plot_color_code_variables()
log_y	boolean log_scale for y axis
limit	integer limit the number of variables to be plotted, Default: 12

Value

grid can be printed with gridExtra::grid.arrange()

See Also

arrangeGrob

Examples

## Not run: 

 # single output example ---------------------------------------
.f                  = randomForest::randomForest
data_ls             = f_clean_data(mtcars)
data                = data_ls$data
formula             = disp~mpg+cyl+am+hp+drat+qsec+vs+gear+carb
m                   = .f(formula, data)
variables           = f_model_importance( m, data)
title               = unlist( stringr::str_split( class(m)[1], '\\.') )[1]
variable_color_code = f_plot_color_code_variables(data_ls)
limit               = 10
log_y               = F

range_variable_num  = data_ls$numericals[1]
range_variable_cat  = data_ls$categoricals[1]

grid_num = f_model_plot_var_dep_over_spec_var_range(m
                                                    , title
                                                    , variables
                                                    , range_variable_num
                                                    , data
                                                    , formula
                                                    , data_ls
                                                    , variable_color_code
                                                    , log_y
                                                    , limit  )
gridExtra::grid.arrange(grid_num)

# pipe example ------------------------------------------------

data_ls = f_clean_data(mtcars)
form = as.formula('disp~cyl+mpg+hp+am+gear+drat+wt+vs+carb')
variable_color_code = f_plot_color_code_variables(data_ls)

grids = pipelearner::pipelearner(data_ls$data) %>%
  pipelearner::learn_models( rpart::rpart, form ) %>%
  pipelearner::learn_models( randomForest::randomForest, form ) %>%
  pipelearner::learn_models( e1071::svm, form ) %>%
  pipelearner::learn() %>%
  dplyr::mutate( imp = map2(fit, train, f_model_importance)
                 , range_var = map_chr(imp, function(x) head(x,1)$row_names )
                 , grid = pmap( list( m = fit
                                      , title = model
                                      , variables = imp
                                      , range_variable = range_var
                                      , data = test
                 )
                 , f_model_plot_var_dep_over_spec_var_range
                 , formula = form
                 , data_ls = data_ls
                 , variable_color_code = variable_color_code
                 , log_y = F
                 , limit = 12
                 )
  )  %>%
  .$grid

f_plot_obj_2_html( grids,  type = "grids", output_file =  'test_me', title = 'Grids', height = 30 )

file.remove('test_me.html')

## End(Not run)

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