Number of randomly sampled predictors

The number of predictors that will be randomly sampled at each split when creating tree models.

Usage

mtry(range = c(1L, unknown()), trans = NULL)

mtry_long(range = c(0L, unknown()), trans = transform_log10())

Arguments

range

A two-element vector holding the defaults for the smallest and largest possible values, respectively. If a transformation is specified, these values should be in the transformed units.

trans

A trans object from the scales package, such as scales::transform_log10() or scales::transform_reciprocal(). If not provided, the default is used which matches the units used in range. If no transformation, NULL.

Details

This parameter is used for regularized or penalized models such as parsnip::rand_forest() and others. mtry_long() has the values on the log10 scale and is helpful when the data contain a large number of predictors.

Since the scale of the parameter depends on the number of columns in the data set, the upper bound is set to unknown but can be filled in via the finalize() method.

Interpretation

mtry_prop() is a variation on mtry() where the value is interpreted as the proportion of predictors that will be randomly sampled at each split rather than the count.

This parameter is not intended for use in accommodating engines that take in this argument as a proportion; mtry is often a main model argument rather than an engine-specific argument, and thus should not have an engine-specific interface.

When wrapping modeling engines that interpret mtry in its sense as a proportion, use the mtry() parameter in parsnip::set_model_arg() and process the passed argument in an internal wrapping function as mtry / number_of_predictors. In addition, introduce a logical argument counts to the wrapping function, defaulting to TRUE, that indicates whether to interpret the supplied argument as a count rather than a proportion.

For an example implementation, see parsnip::xgb_train().

See also

mtry_prop

Examples

mtry(c(1L, 10L)) # in original units
#> # Randomly Selected Predictors (quantitative)
#> Range: [1, 10]
mtry_long(c(0, 5)) # in log10 units
#> # Randomly Selected Predictors (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [0, 5]