setup: Setup your modeling workflow

In realgabon/insuRglm: Tools for GLM Modeling in Insurance Context

Setup your modeling workflow

Description

Creates a setup object that is the basis for any insuRglm modeling workflow. This object is subsequently used as a main input in most functions in the package.

Usage

setup(
  data_train,
  data_test = NULL,
  target,
  weight = NULL,
  offset = NULL,
  family = c("poisson", "gamma", "tweedie"),
  tweedie_p = NULL,
  simple_factors = NULL,
  keep_cols = NULL,
  glm_backend = c("speedglm", "stats"),
  folder = getwd(),
  load_file_nm = NULL,
  save_file_nm = NULL,
  seed = NULL
)

Arguments

data_train	Dataframe. Training data
data_test	Dataframe. Test data
target	Character scalar. Name of the target variable
weight	Character scalar. Name of the weight variable
family	Character scalar. Name of distribution family. One of poisson, tweedie or gamma
tweedie_p	Numeric scalar. Tweedie variance power, if family tweedie is used
simple_factors	Character vector. Names of potential predictors. These predictors need to be factor class.
keep_cols	Character vector. Names of columns that are not potential predictors, but should be kept in data.
glm_backend	Character scalar. Either 'speedglm' or 'stats'. Choosing 'speedglm' results in using speedglm::speedglm as glm backend, while choosing 'stats' will result in traditional stats::glm.
folder	Character scalar. Path to an existing folder where setup/model files will be stored.
load_file_nm	Character scalar. Filename of an existing setup object created by running setup. Must be within folder specified by folder. Can be without the '_setup.rds' suffix.
save_file_nm	Character scalar. Filename of a setup object saved during this run of the setup function. Will be saved within the folder specified by folder,
seed	Numeric scalar. Seed for reproducible random number generation, e.g. for creating CV folds.
offset.	Character scalar. Name of the offset variable, applicable for poisson family

Value

List of class setup. Contains attributes and objects used by other functions in the package.

Note

Short summary of the train/test datasets is written to the console

Examples

require(dplyr) # for the pipe operator#'

# poisson distribution target
data('freq_train')

setup <- setup(
  data_train = freq_train,
  target = 'freq',
  offset = 'exposure',
  family = 'poisson',
  keep_cols = c('pol_nbr', 'premium')
)

# gamma distribution target
data('sev_train')

setup <- setup(
  data_train = sev_train,
  target = 'sev',
  weight = 'numclaims',
  family = 'gamma',
  keep_cols = c('pol_nbr', 'exposure', 'premium')
)

# tweedie distribution - burning cost
data('bc_train')

setup <- setup(
  data_train = bc_train,
  target = 'bc',
  weight = 'exposure',
  family = 'tweedie',
  tweedie_p = 1.75, # use tweedie::tweedie.profile to determine the best value
  keep_cols = c('pol_nbr', 'premium')
)

# tweedie distribution - loss ratio
data('lr_train')

setup <- setup(
  data_train = lr_train,
  target = 'lr',
  weight = 'premium',
  family = 'tweedie',
  tweedie_p = 1.75, # use tweedie::tweedie.profile to determine the best value
  keep_cols = c('pol_nbr', 'exposure')
)

realgabon/insuRglm documentation built on Jan. 2, 2023, 2:51 a.m.