f_predict_plot_regression_alluvials: plot residuals of different models as aaluvials

In erblast/oetteR: Collection of personal R functions

Description

Residuals will be binned using boxplotstats with the modification, that the median will be set to zero. If quantile boundaries do not fall into sensible ranges they will be replaced by the standard error (SE). For example if we have the following boxplotstats -2, 1, 3, 5, 8 with an SE of 0.5 we will modify the boundaries as following. -2, -0.5, 0, 3, 5 . The reason is that we want to be able to follow observations with positive or negative residuals through the alluvial plot in order to judge emerging patterns. The models will be sorted by MSE and the alluvial plot will be flipped with the model with the lowest MSE on top.

Usage

f_predict_plot_regression_alluvials(data, col_id, col_title = "title",
  col_pred = "pred", col_obs = "target1", ...)

Arguments

data	dataframe
col_id	character vecotr dentoing id column
col_title	character vector denoting model title column, Default: 'title'
col_pred	character vector denoting prediciont column, Default: 'pred'
col_obs	character vector denoting column with observed values, Default: 'target1'
...	additional arguments passed to f_plot_alluvial_1v1

Value

plot

See Also

brewer.pal

Examples

  form = as.formula( 'displacement~cylinders+mpg')

df = ISLR::Auto %>%
  mutate( name = paste( name, row_number() ) ) %>%
  pipelearner::pipelearner() %>%
  pipelearner::learn_models( rpart::rpart, form ) %>%
  pipelearner::learn_models( randomForest::randomForest, form ) %>%
  pipelearner::learn_models( e1071::svm, form ) %>%
  pipelearner::learn() %>%
  f_predict_pl_regression( 'name' ) %>%
  unnest(preds)

f_predict_plot_regression_alluvials(df
                                    , col_id = 'name'
                                    , col_title = 'model'
                                    , col_pred = 'pred'
                                    , col_obs = 'target1')

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