cross_entropy_control: Hyperparmaeters for the cross-entropy models

In marcusbuckmann/ffcr: Train two families of transparent classification models: fast-and-frugal trees and tallying

Hyperparmaeters for the cross-entropy models

Description

cross_entropy_control Returns a list of parameters that control the cross-entropy optimization procedure. The output of this function is be passed when training a fast-and-frugal tree or tallying model using the cross-entropy method (see examples).

Usage

cross_entropy_control(
  starts = 10,
  learning_rate = 0.05,
  maximum_time = 3600,
  iterations = 500,
  early_stopping = 25,
  thresholds = 100,
  split_percentiles = FALSE,
  samples = 100,
  elite_samples = 5,
  threads = parallel::detectCores() - 1
)

Arguments

starts	Number of fast-and-frugal trees or tallying models generated with different seeds. The best-fitting model will be returned. The default is 10.
learning_rate	Learning rate used in the optimization procedure. The higher the value, the quicker the model will converge. However, at higher learning rates, the model is more likely to converge to less accurate fast-and-frugal trees and tallying models. The default value is 0.1.
maximum_time	Maximum training time (in seconds) to obtain the final model.
iterations	The maximum number of iterations. The default is 500.
early_stopping	Training stops early if the performance does not improve after this many iterations in a row. The default is 25.
thresholds	The maximum number of numeric thresholds that are tested for a numeric feature. The default is 100.
split_percentiles	When TRUE, the candidate thresholds at which features can be split are the percentiles of the features' distribution. When FALSE the candidate values are obtained by dividing a feature's values in equidistant bins. The default value is FALSE.
samples	Number of models created in each iteration. The default is 100.
elite_samples	The number of best-performing models that are used to update the parameter distribution. The default is 10.
threads	Number of CPU cores used. The default is the number of available cores minus 1.

Examples

## Not run: 
data(liver)
liver$sex <- ifelse(liver$sex == "Female", 1,0) # Recoding categorical feature because the cross-entropy method only works with numeric features.
model_tree <- fftree(data = liver, formula = diagnosis~., method = "cross-entropy", cross_entropy_parameters = cross_entropy_control(starts = 5))
model_tally <- tally(data = liver, formula = diagnosis~., method = "cross-entropy", cross_entropy_parameters = cross_entropy_control(starts = 5))

## End(Not run)

marcusbuckmann/ffcr documentation built on Jan. 4, 2024, 3:45 p.m.