spect_train: Generates a trained caret model using the given primary binary classification. Optionally generates a stacked ensemble model if a list of base learners is supplied.

Description

Generates a trained caret model using the given primary binary classification. Optionally generates a stacked ensemble model if a list of base learners is supplied.

Usage

spect_train(
  test_prop = 0.2,
  censor_type = "half",
  bin_slices = 10,
  method = "repeatedcv",
  resampling_number = 10,
  kfold_repeats = 3,
  model_algorithm,
  base_learner_list = list(),
  metric = "Kappa",
  rng_seed = 42,
  use_parallel = TRUE,
  cores = 0,
  modeling_data,
  event_indicator_var,
  survival_time_var,
  obs_window
)

Value

A list containing all intermediate data sets created by `spect_train`, a trained caret model object, the following parameters passed to `spect_train`: `obs_window`, `survival_time_var`, `event_indicator_var`, `base_learner_list`, `bin_slices`, and the bounds of each interval generated by the training data set.

Arguments

test_prop: optional proportion of the data set to reserve for testing
censor_type: optional method used to determine censorship in a given bin - may be "half", "prev" or "same". see createDiscreteDat for usage.
bin_slices: optional number of intervals to use for predictions.
method: optional caret parameter
resampling_number: optional for repeated cv
kfold_repeats: optional number of folds
model_algorithm: primary classification algorithm. Trains a stack-ensemble model if `base_learner_list` is supplied, otherwise trains a simple classifier model.
base_learner_list: optional list of base learner algorithms
metric: optional metric for model calibration
rng_seed: optional random number generation seed for reproducibility
use_parallel: optioanlly make use of the caret multicore training cluster
cores: optioanl number of cores for multicore training. If zero, spect will attempt to make a good choice. Note: only relevant if `use_parallel` is set to TRUE, otherwise this parameter is ignored.
modeling_data: This data set must have one column for time and one column for the event indicator. The remaining columns are treated as covariates for modeling.
event_indicator_var: The name of the column containing the event indicator (values in this column must be zero or one).
survival_time_var: The name of the column containing the time variable
obs_window: The last time to use for generating person-period data. Any event occurring after this time will be administratively censored. In general, choosing a time at or near the end of the max observed time will include most events.

Author

Stephen Abrams, stephen.abrams@louisville.edu